xuning / sherpaonnx

转到一个项目
Toggle navigation Toggle navigation pinning
Fangjun Kuang
Committed by GitHub
e63d205b 1 parent 6125d9e0

Add go-api-examples (#219)

内嵌 并排对比
正在显示 28 个修改的文件 包含 1624 行增加 和 14 行删除
name: test-go
on:
push:
branches:
- master
tags:
- '*'
pull_request:
branches:
- master
workflow_dispatch:
concurrency:
group: go-${{ github.ref }}
cancel-in-progress: true
jobs:
go:
name: go ${{ matrix.os }} ${{matrix.arch }}
runs-on: ${{ matrix.os }}
strategy:
fail-fast: false
matrix:
include:
- os: ubuntu-latest
arch: amd64
- os: macos-latest
arch: amd64
- os: windows-latest
arch: x64
- os: windows-latest
arch: x86 # use 386 for GOARCH
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- uses: actions/setup-go@v4
with:
go-version: '>=1.20'
- name: Display go version
shell: bash
run: |
go version
go env GOPATH
go env GOARCH
- name: Set up MinGW
if: matrix.os == 'windows-latest'
uses: egor-tensin/setup-mingw@v2
with:
platform: ${{ matrix.arch }}
- name: Show gcc
if: matrix.os == 'windows-latest'
run: |
gcc --version
- name: Test non-streaming decoding files (Linux/macOS)
if: matrix.os != 'windows-latest'
shell: bash
run: |
cd go-api-examples/non-streaming-decode-files
ls -lh
go mod tidy
cat go.mod
go build -x
ls -lh
git lfs install
echo "Test transducer"
git clone https://huggingface.co/csukuangfj/sherpa-onnx-zipformer-en-2023-06-26
./run-transducer.sh
rm -rf sherpa-onnx-zipformer-en-2023-06-26
echo "Test paraformer"
git clone https://huggingface.co/csukuangfj/sherpa-onnx-paraformer-zh-2023-03-28
./run-paraformer.sh
rm -rf sherpa-onnx-paraformer-zh-2023-03-28
echo "Test NeMo CTC"
git clone https://huggingface.co/csukuangfj/sherpa-onnx-nemo-ctc-en-conformer-medium
./run-nemo-ctc.sh
rm -rf sherpa-onnx-nemo-ctc-en-conformer-medium
- name: Test non-streaming decoding files (Win64)
if: matrix.os == 'windows-latest' && matrix.arch == 'x64'
shell: bash
run: |
cd go-api-examples/non-streaming-decode-files
ls -lh
go mod tidy
cat go.mod
go build
ls -lh
echo $PWD
ls -lh /C/Users/runneradmin/go/pkg/mod/github.com/k2-fsa/
ls -lh /C/Users/runneradmin/go/pkg/mod/github.com/k2-fsa/*
cp -v /C/Users/runneradmin/go/pkg/mod/github.com/k2-fsa/sherpa-onnx-go-windows*/lib/x86_64-pc-windows-gnu/*.dll .
ls -lh
git lfs install
echo "Test transducer"
git clone https://huggingface.co/csukuangfj/sherpa-onnx-zipformer-en-2023-06-26
./run-transducer.sh
rm -rf sherpa-onnx-zipformer-en-2023-06-26
echo "Test paraformer"
git clone https://huggingface.co/csukuangfj/sherpa-onnx-paraformer-zh-2023-03-28
./run-paraformer.sh
rm -rf sherpa-onnx-paraformer-zh-2023-03-28
echo "Test NeMo CTC"
git clone https://huggingface.co/csukuangfj/sherpa-onnx-nemo-ctc-en-conformer-medium
./run-nemo-ctc.sh
rm -rf sherpa-onnx-nemo-ctc-en-conformer-medium
- name: Test non-streaming decoding files (Win32)
if: matrix.os == 'windows-latest' && matrix.arch == 'x86'
shell: bash
run: |
cd go-api-examples/non-streaming-decode-files
ls -lh
go mod tidy
cat go.mod
ls -lh
go env GOARCH
go env
echo "------------------------------"
go env -w GOARCH=386
go env -w CGO_ENABLED=1
go env
go clean
go build -x
echo $PWD
ls -lh /C/Users/runneradmin/go/pkg/mod/github.com/k2-fsa/
cp -v /C/Users/runneradmin/go/pkg/mod/github.com/k2-fsa/sherpa-onnx-go-windows*/lib/i686-pc-windows-gnu/*.dll .
ls -lh
git lfs install
echo "Test transducer"
git clone https://huggingface.co/csukuangfj/sherpa-onnx-zipformer-en-2023-06-26
./run-transducer.sh
rm -rf sherpa-onnx-zipformer-en-2023-06-26
echo "Test paraformer"
git clone https://huggingface.co/csukuangfj/sherpa-onnx-paraformer-zh-2023-03-28
./run-paraformer.sh
rm -rf sherpa-onnx-paraformer-zh-2023-03-28
echo "Test NeMo CTC"
git clone https://huggingface.co/csukuangfj/sherpa-onnx-nemo-ctc-en-conformer-medium
./run-nemo-ctc.sh
rm -rf sherpa-onnx-nemo-ctc-en-conformer-medium
- name: Test streaming decoding files (Linux/macOS)
if: matrix.os != 'windows-latest'
shell: bash
run: |
cd go-api-examples/streaming-decode-files
ls -lh
go mod tidy
cat go.mod
go build -x
ls -lh
git lfs install
echo "Test transducer"
git clone https://huggingface.co/csukuangfj/sherpa-onnx-streaming-zipformer-en-2023-06-26
./run.sh
rm -rf sherpa-onnx-streaming-zipformer-en-2023-06-26
- name: Test streaming decoding files (Win64)
if: matrix.os == 'windows-latest' && matrix.arch == 'x64'
shell: bash
run: |
cd go-api-examples/streaming-decode-files
ls -lh
go mod tidy
cat go.mod
go build
ls -lh
echo $PWD
ls -lh /C/Users/runneradmin/go/pkg/mod/github.com/k2-fsa/
ls -lh /C/Users/runneradmin/go/pkg/mod/github.com/k2-fsa/*
cp -v /C/Users/runneradmin/go/pkg/mod/github.com/k2-fsa/sherpa-onnx-go-windows*/lib/x86_64-pc-windows-gnu/*.dll .
ls -lh
git lfs install
echo "Test transducer"
git clone https://huggingface.co/csukuangfj/sherpa-onnx-streaming-zipformer-en-2023-06-26
./run.sh
rm -rf sherpa-onnx-streaming-zipformer-en-2023-06-26
- name: Test streaming decoding files (Win32)
if: matrix.os == 'windows-latest' && matrix.arch == 'x86'
shell: bash
run: |
cd go-api-examples/streaming-decode-files
ls -lh
go mod tidy
cat go.mod
ls -lh
go env GOARCH
go env
echo "------------------------------"
go env -w GOARCH=386
go env -w CGO_ENABLED=1
go env
go clean
go build -x
echo $PWD
ls -lh /C/Users/runneradmin/go/pkg/mod/github.com/k2-fsa/
ls -lh /C/Users/runneradmin/go/pkg/mod/github.com/k2-fsa/*
cp -v /C/Users/runneradmin/go/pkg/mod/github.com/k2-fsa/sherpa-onnx-go-windows*/lib/i686-pc-windows-gnu/*.dll .
ls -lh
git lfs install
echo "Test transducer"
git clone https://huggingface.co/csukuangfj/sherpa-onnx-streaming-zipformer-en-2023-06-26
./run.sh
rm -rf sherpa-onnx-streaming-zipformer-en-2023-06-26
... ...
name: release-go
on:
push:
branches:
- master
tags:
- '*'
workflow_dispatch:
inputs:
version:
description: "Version information（e.g., 1.5.3) or auto"
required: true
env:
VERSION:
|- # Enter release tag name or version name in workflow_dispatch. Appropriate version if not specified
${{ github.event.release.tag_name || github.event.inputs.version }}
concurrency:
group: release-go-${{ github.ref }}
cancel-in-progress: true
jobs:
linux-x86_64_wheel:
if: github.repository_owner == 'k2-fsa' || github.repository_owner == 'csukuangfj'
name: Linux x86_64
runs-on: ${{ matrix.os }}
strategy:
fail-fast: false
matrix:
os: [ubuntu-latest]
steps:
- uses: actions/checkout@v2
- name: SSH to GitHub
run: |
mkdir -p ~/.ssh/
cp scripts/go/ssh_config ~/.ssh/config
echo "${{ secrets.MY_GITHUB_SSH_KEY }}" > ~/.ssh/github && chmod 600 ~/.ssh/github
ssh github.com || true
rm ~/.ssh/github
# see https://cibuildwheel.readthedocs.io/en/stable/changelog/
# for a list of versions
- name: Build wheels
uses: pypa/cibuildwheel@v2.11.4
env:
CIBW_BEFORE_BUILD: "pip install -U cmake numpy"
CIBW_BUILD: "cp38-*64"
CIBW_SKIP: "cp27-* cp35-* cp36-* *-win32 pp* *-musllinux* *-manylinux_i686"
CIBW_BUILD_VERBOSITY: 3
CIBW_ENVIRONMENT_LINUX: LD_LIBRARY_PATH='/project/build/bdist.linux-x86_64/wheel/sherpa_onnx/lib'
- name: Display wheels
shell: bash
run: |
ls -lh ./wheelhouse/*.whl
unzip -l ./wheelhouse/*.whl
- uses: actions/upload-artifact@v2
with:
name: ${{ matrix.os }}-wheels-for-go
path: ./wheelhouse/*.whl
macOS:
name: macOS ${{ matrix.arch }}
runs-on: ${{ matrix.os }}
strategy:
fail-fast: false
matrix:
os: [macos-latest]
arch: [x86_64, arm64]
steps:
- uses: actions/checkout@v2
- name: Configure CMake
shell: bash
run: |
mkdir build
cd build
cmake -D CMAKE_BUILD_TYPE=${{ matrix.build_type }} -D BUILD_SHARED_LIBS=ON -DCMAKE_OSX_ARCHITECTURES=${{ matrix.arch }} -DCMAKE_INSTALL_PREFIX=./install ..
- name: Build sherpa-onnx for macOS ${{ matrix.arch }}
shell: bash
run: |
cd build
make -j2
make install
ls -lh lib
ls -lh bin
file install/lib/lib*
- uses: actions/upload-artifact@v2
with:
name: ${{ matrix.os }}-for-${{ matrix.arch }}
path: ./build/install/lib/
windows:
name: Windows ${{ matrix.arch }}
runs-on: ${{ matrix.os }}
strategy:
fail-fast: false
matrix:
os: [windows-latest]
arch: [x64, Win32]
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Configure CMake
shell: bash
run: |
mkdir build
cd build
cmake -A ${{ matrix.arch }} -D CMAKE_BUILD_TYPE=Release -D BUILD_SHARED_LIBS=ON -DCMAKE_INSTALL_PREFIX=./install ..
- name: Build sherpa-onnx for windows
shell: bash
run: |
cd build
cmake --build . --config Release -- -m:2
cmake --build . --config Release --target install -- -m:2
ls -lh install/*
ls -lh install/lib
ls -lh install/bin
- name: Upload artifact
uses: actions/upload-artifact@v2
with:
name: sherpa-onnx-go-windows-${{ matrix.arch }}
path: ./build/install/lib/
Release:
name: Release
runs-on: ubuntu-latest
needs: [linux-x86_64_wheel, macOS, windows]
steps:
- uses: actions/checkout@v2
- name: Add SSH key
run: |
mkdir -p ~/.ssh/
cp scripts/go/ssh_config ~/.ssh/config
echo "${{ secrets.MY_GITHUB_SSH_KEY }}" > ~/.ssh/github && chmod 600 ~/.ssh/github
ssh github.com || true
- name: Retrieve artifact from ubuntu-latest
uses: actions/download-artifact@v2
with:
name: ubuntu-latest-wheels-for-go
path: ./linux
- name: Retrieve artifact from macos-latest (x86_64)
uses: actions/download-artifact@v2
with:
name: macos-latest-for-x86_64
path: ./macos-x86_64
- name: Retrieve artifact from macos-latest (arm64)
uses: actions/download-artifact@v2
with:
name: macos-latest-for-arm64
path: ./macos-arm64
- name: Retrieve artifact from windows-latest (x64)
uses: actions/download-artifact@v2
with:
name: sherpa-onnx-go-windows-x64
path: ./windows-x64
- name: Retrieve artifact from windows-latest (Win32)
uses: actions/download-artifact@v2
with:
name: sherpa-onnx-go-windows-Win32
path: ./windows-win32
- name: Unzip Ubuntu wheels
shell: bash
run: |
cd linux
ls -lh
unzip ./*.whl
tree .
- name: Release go
if: env.VERSION != ''
shell: bash
run: |
./scripts/go/release.sh
... ...
... ... @@ -28,7 +28,7 @@ jobs:
fail-fast: false
matrix:
os: [ubuntu-latest, windows-latest, macos-latest]
python-version: ["3.7", "3.8", "3.9", "3.10"]
python-version: ["3.8", "3.9", "3.10"]
steps:
- uses: actions/checkout@v2
... ...
... ... @@ -8,10 +8,6 @@ if(NOT CMAKE_SYSTEM_NAME STREQUAL Darwin)
message(FATAL_ERROR "This file is for macOS only. Given: ${CMAKE_SYSTEM_NAME}")
endif()
if(NOT CMAKE_SYSTEM_PROCESSOR STREQUAL arm64)
message(FATAL_ERROR "This file is for arm64 only. Given: ${CMAKE_SYSTEM_PROCESSOR}")
endif()
set(onnxruntime_URL "https://github.com/microsoft/onnxruntime/releases/download/v1.15.1/onnxruntime-osx-arm64-1.15.1.tgz")
set(onnxruntime_URL2 "https://huggingface.co/csukuangfj/sherpa-onnx-cmake-deps/resolve/main/onnxruntime-osx-arm64-1.15.1.tgz")
set(onnxruntime_HASH "SHA256=df97832fc7907c6677a6da437f92339d84a462becb74b1d65217fcb859ee9460")
... ...
... ... @@ -8,10 +8,6 @@ if(NOT CMAKE_SYSTEM_NAME STREQUAL Darwin)
message(FATAL_ERROR "This file is for macOS only. Given: ${CMAKE_SYSTEM_NAME}")
endif()
if(NOT CMAKE_SYSTEM_PROCESSOR STREQUAL x86_64)
message(FATAL_ERROR "This file is for x86_64 only. Given: ${CMAKE_SYSTEM_PROCESSOR}")
endif()
set(onnxruntime_URL "https://github.com/microsoft/onnxruntime/releases/download/v1.15.1/onnxruntime-osx-x86_64-1.15.1.tgz")
set(onnxruntime_URL2 "https://huggingface.co/csukuangfj/sherpa-onnx-cmake-deps/resolve/main/onnxruntime-osx-x86_64-1.15.1.tgz")
set(onnxruntime_HASH "SHA256=4b66ebbca24b8b96f6b74655fee3610a7e529b4e01f6790632f24ee82b778e5a")
... ...
... ... @@ -16,12 +16,18 @@ function(download_onnxruntime)
include(onnxruntime-linux-x86_64)
endif()
elseif(CMAKE_SYSTEM_NAME STREQUAL Darwin)
if (arm64 IN_LIST CMAKE_OSX_ARCHITECTURES OR x86_64 IN_LIST CMAKE_OSX_ARCHITECTURES)
include(onnxruntime-darwin-universal)
elseif(CMAKE_SYSTEM_PROCESSOR STREQUAL x86_64)
include(onnxruntime-darwin-x86_64)
if (arm64 IN_LIST CMAKE_OSX_ARCHITECTURES AND x86_64 IN_LIST CMAKE_OSX_ARCHITECTURES)
include(onnxruntime-osx-universal)
elseif(CMAKE_SYSTEM_PROCESSOR STREQUAL x86_64 AND CMAKE_OSX_ARCHITECTURES STREQUAL "arm64")
# cross compiling
include(onnxruntime-osx-arm64)
elseif(CMAKE_SYSTEM_PROCESSOR STREQUAL arm64 AND CMAKE_OSX_ARCHITECTURES STREQUAL "x86_64")
# cross compiling
include(onnxruntime-osx-x86_64)
elseif(CMAKE_SYSTEM_PROCESSOR STREQUAL arm64)
include(onnxruntime-darwin-arm64)
include(onnxruntime-osx-arm64)
elseif(CMAKE_SYSTEM_PROCESSOR STREQUAL x86_64)
include(onnxruntime-osx-x86_64)
else()
message(FATAL_ERROR "Unsupport processor {CMAKE_SYSTEM_PROCESSOR} for Darwin")
endif()
... ...
non-streaming-decode-files
sherpa-onnx-zipformer-en-2023-06-26
... ...
module non-streaming-decode-files
go 1.20
require (
github.com/k2-fsa/sherpa-onnx-go v1.5.3-alpha.8
github.com/spf13/pflag v1.0.5
github.com/youpy/go-wav v0.3.2
)
require (
github.com/k2-fsa/sherpa-onnx-go-linux v1.5.3-alpha.6 // indirect
github.com/k2-fsa/sherpa-onnx-go-macos v1.5.3-alpha.4 // indirect
github.com/k2-fsa/sherpa-onnx-go-windows v1.5.3-alpha.5 // indirect
github.com/youpy/go-riff v0.1.0 // indirect
github.com/zaf/g711 v0.0.0-20190814101024-76a4a538f52b // indirect
)
... ...
github.com/davecgh/go-spew v1.1.0 h1:ZDRjVQ15GmhC3fiQ8ni8+OwkZQO4DARzQgrnXU1Liz8=
github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
github.com/google/go-cmp v0.5.6 h1:BKbKCqvP6I+rmFHt06ZmyQtvB8xAkWdhFyr0ZUNZcxQ=
github.com/google/go-cmp v0.5.6/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE=
github.com/k2-fsa/sherpa-onnx-go v1.5.3-alpha.8 h1:BXc31pWwd7CJLM6m9HavxqiyYJdN3Jc9I26pd4x+JHE=
github.com/k2-fsa/sherpa-onnx-go v1.5.3-alpha.8/go.mod h1:kszL/pwg9XTpRGi1AYW/aSwdhRBqb6LN0SjXR0jnsBo=
github.com/k2-fsa/sherpa-onnx-go-linux v1.5.3-alpha.6 h1:o0+l4Wr3IWkWH+kdt8ZZP55L8mRSpW7h1KzvXcfx9FM=
github.com/k2-fsa/sherpa-onnx-go-linux v1.5.3-alpha.6/go.mod h1:9kU02PSdDdzBApwIDBmE2jWS54WvZbRafOEvW/PVLxE=
github.com/k2-fsa/sherpa-onnx-go-macos v1.5.3-alpha.4 h1:Xri23R3+tQFkNwO6zVxqZbjMRJP40z7JtoSqW6UP6sM=
github.com/k2-fsa/sherpa-onnx-go-macos v1.5.3-alpha.4/go.mod h1:a+AJZKNQkFO+JyzGkHySysYfBzzdcoJI5ITFsnhVcmo=
github.com/k2-fsa/sherpa-onnx-go-windows v1.5.3-alpha.5 h1:OR1LLoptR8W35j5u06KrYUblG35B83HHCPWrVK31uHY=
github.com/k2-fsa/sherpa-onnx-go-windows v1.5.3-alpha.5/go.mod h1:ermMOETZUv0nM7MmnXWqeHREMR6zQXBhJpEP2fbHIZo=
github.com/pkg/errors v0.9.1 h1:FEBLx1zS214owpjy7qsBeixbURkuhQAwrK5UwLGTwt4=
github.com/pkg/errors v0.9.1/go.mod h1:bwawxfHBFNV+L2hUp1rHADufV3IMtnDRdf1r5NINEl0=
github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=
github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
github.com/spf13/pflag v1.0.5 h1:iy+VFUOCP1a+8yFto/drg2CJ5u0yRoB7fZw3DKv/JXA=
github.com/spf13/pflag v1.0.5/go.mod h1:McXfInJRrz4CZXVZOBLb0bTZqETkiAhM9Iw0y3An2Bg=
github.com/stretchr/objx v0.1.0/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+wExME=
github.com/stretchr/testify v1.6.1 h1:hDPOHmpOpP40lSULcqw7IrRb/u7w6RpDC9399XyoNd0=
github.com/stretchr/testify v1.6.1/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/h/Wwjteg=
github.com/youpy/go-riff v0.1.0 h1:vZO/37nI4tIET8tQI0Qn0Y79qQh99aEpponTPiPut7k=
github.com/youpy/go-riff v0.1.0/go.mod h1:83nxdDV4Z9RzrTut9losK7ve4hUnxUR8ASSz4BsKXwQ=
github.com/youpy/go-wav v0.3.2 h1:NLM8L/7yZ0Bntadw/0h95OyUsen+DQIVf9gay+SUsMU=
github.com/youpy/go-wav v0.3.2/go.mod h1:0FCieAXAeSdcxFfwLpRuEo0PFmAoc+8NU34h7TUvk50=
github.com/zaf/g711 v0.0.0-20190814101024-76a4a538f52b h1:QqixIpc5WFIqTLxB3Hq8qs0qImAgBdq0p6rq2Qdl634=
github.com/zaf/g711 v0.0.0-20190814101024-76a4a538f52b/go.mod h1:T2h1zV50R/q0CVYnsQOQ6L7P4a2ZxH47ixWcMXFGyx8=
golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c h1:dUUwHk2QECo/6vqA44rthZ8ie2QXMNeKRTHCNY2nXvo=
gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
gotest.tools v2.2.0+incompatible h1:VsBPFP1AI068pPrMxtb/S8Zkgf9xEmTLJjfM+P5UIEo=
gotest.tools v2.2.0+incompatible/go.mod h1:DsYFclhRJ6vuDpmuTbkuFWG+y2sxOXAzmJt81HFBacw=
... ...
package main
import (
"bytes"
"encoding/binary"
sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
flag "github.com/spf13/pflag"
"github.com/youpy/go-wav"
"os"
"strings"
"log"
)
func main() {
log.SetFlags(log.LstdFlags | log.Lmicroseconds)
config := sherpa.OfflineRecognizerConfig{}
config.FeatConfig = sherpa.FeatureConfig{SampleRate: 16000, FeatureDim: 80}
flag.StringVar(&config.ModelConfig.Transducer.Encoder, "encoder", "", "Path to the encoder model")
flag.StringVar(&config.ModelConfig.Transducer.Decoder, "decoder", "", "Path to the decoder model")
flag.StringVar(&config.ModelConfig.Transducer.Joiner, "joiner", "", "Path to the joiner model")
flag.StringVar(&config.ModelConfig.Paraformer.Model, "paraformer", "", "Path to the paraformer model")
flag.StringVar(&config.ModelConfig.NemoCTC.Model, "nemo-ctc", "", "Path to the NeMo CTC model")
flag.StringVar(&config.ModelConfig.Tokens, "tokens", "", "Path to the tokens file")
flag.IntVar(&config.ModelConfig.NumThreads, "num-threads", 1, "Number of threads for computing")
flag.IntVar(&config.ModelConfig.Debug, "debug", 0, "Whether to show debug message")
flag.StringVar(&config.ModelConfig.ModelType, "model-type", "", "Optional. Used for loading the model in a faster way")
flag.StringVar(&config.ModelConfig.Provider, "provider", "cpu", "Provider to use")
flag.StringVar(&config.LmConfig.Model, "lm-model", "", "Optional. Path to the LM model")
flag.Float32Var(&config.LmConfig.Scale, "lm-scale", 1.0, "Optional. Scale for the LM model")
flag.StringVar(&config.DecodingMethod, "decoding-method", "greedy_search", "Decoding method. Possible values: greedy_search, modified_beam_search")
flag.IntVar(&config.MaxActivePaths, "max-active-paths", 4, "Used only when --decoding-method is modified_beam_search")
flag.Parse()
if len(flag.Args()) != 1 {
log.Fatalf("Please provide one wave file")
}
log.Println("Reading", flag.Arg(0))
samples, sampleRate := readWave(flag.Arg(0))
log.Println("Initializing recognizer (may take several seconds)")
recognizer := sherpa.NewOfflineRecognizer(&config)
log.Println("Recognizer created!")
defer sherpa.DeleteOfflineRecognizer(recognizer)
log.Println("Start decoding!")
stream := sherpa.NewOfflineStream(recognizer)
defer sherpa.DeleteOfflineStream(stream)
stream.AcceptWaveform(sampleRate, samples)
recognizer.Decode(stream)
log.Println("Decoding done!")
result := stream.GetResult()
log.Println(strings.ToLower(result.Text))
log.Printf("Wave duration: %v seconds", float32(len(samples))/float32(sampleRate))
}
func readWave(filename string) (samples []float32, sampleRate int) {
file, _ := os.Open(filename)
defer file.Close()
reader := wav.NewReader(file)
format, err := reader.Format()
if err != nil {
log.Fatalf("Failed to read wave format")
}
if format.AudioFormat != 1 {
log.Fatalf("Support only PCM format. Given: %v\n", format.AudioFormat)
}
if format.NumChannels != 1 {
log.Fatalf("Support only 1 channel wave file. Given: %v\n", format.NumChannels)
}
if format.BitsPerSample != 16 {
log.Fatalf("Support only 16-bit per sample. Given: %v\n", format.BitsPerSample)
}
reader.Duration() // so that it initializes reader.Size
buf := make([]byte, reader.Size)
n, err := reader.Read(buf)
if n != int(reader.Size) {
log.Fatalf("Failed to read %v bytes. Returned %v bytes\n", reader.Size, n)
}
samples = samplesInt16ToFloat(buf)
sampleRate = int(format.SampleRate)
return
}
func samplesInt16ToFloat(inSamples []byte) []float32 {
numSamples := len(inSamples) / 2
outSamples := make([]float32, numSamples)
for i := 0; i != numSamples; i++ {
s := inSamples[i*2 : (i+1)*2]
var s16 int16
buf := bytes.NewReader(s)
err := binary.Read(buf, binary.LittleEndian, &s16)
if err != nil {
log.Fatal("Failed to parse 16-bit sample")
}
outSamples[i] = float32(s16) / 32768
}
return outSamples
}
... ...
#!/usr/bin/env bash
# Please refer to
# https://k2-fsa.github.io/sherpa/onnx/pretrained_models/offline-ctc/nemo/english.html#stt-en-conformer-ctc-medium
# to download the model
# before you run this script.
#
# You can switch to a different online model if you need
./non-streaming-decode-files \
--nemo-ctc ./sherpa-onnx-nemo-ctc-en-conformer-medium/model.onnx \
--tokens ./sherpa-onnx-nemo-ctc-en-conformer-medium/tokens.txt \
--model-type nemo_ctc \
--debug 0 \
./sherpa-onnx-nemo-ctc-en-conformer-medium/test_wavs/0.wav
... ...
#!/usr/bin/env bash
# Please refer to
# https://k2-fsa.github.io/sherpa/onnx/pretrained_models/offline-paraformer/paraformer-models.html#csukuangfj-sherpa-onnx-paraformer-zh-2023-03-28-chinese
# to download the model
# before you run this script.
#
# You can switch to a different online model if you need
./non-streaming-decode-files \
--paraformer ./sherpa-onnx-paraformer-zh-2023-03-28/model.int8.onnx \
--tokens ./sherpa-onnx-paraformer-zh-2023-03-28/tokens.txt \
--model-type paraformer \
--debug 0 \
./sherpa-onnx-paraformer-zh-2023-03-28/test_wavs/0.wav
... ...
#!/usr/bin/env bash
# Please refer to
# https://k2-fsa.github.io/sherpa/onnx/pretrained_models/offline-transducer/zipformer-transducer-models.html#csukuangfj-sherpa-onnx-zipformer-en-2023-06-26-english
# to download the model
# before you run this script.
#
# You can switch to a different online model if you need
./non-streaming-decode-files \
--encoder ./sherpa-onnx-zipformer-en-2023-06-26/encoder-epoch-99-avg-1.onnx \
--decoder ./sherpa-onnx-zipformer-en-2023-06-26/decoder-epoch-99-avg-1.onnx \
--joiner ./sherpa-onnx-zipformer-en-2023-06-26/joiner-epoch-99-avg-1.onnx \
--tokens ./sherpa-onnx-zipformer-en-2023-06-26/tokens.txt \
--model-type transducer \
--debug 0 \
./sherpa-onnx-zipformer-en-2023-06-26/test_wavs/0.wav
... ...
real-time-speech-recognition-from-microphone
... ...
# Introduction
This examples shows how to use the golang package of [sherpa-onnx][sherpa-onnx]
for real-time speech recognition from microphone.
It uses <https://github.com/gordonklaus/portaudio>
to read the microphone and you have to install `portaudio` first.
On macOS, you can use
```
brew install portaudio
```
and it will install `portaudio` into `/usr/local/Cellar/portaudio/19.7.0`.
You need to set the following environment variable
```
export PKG_CONFIG_PATH=/usr/local/Cellar/portaudio/19.7.0
```
so that `pkg-config --cflags --libs portaudio-2.0` can run successfully.
[sherpa-onnx]: https://github.com/k2-fsa/sherpa-onnx
... ...
module real-time-speech-recognition-from-microphone
go 1.20
require (
github.com/gordonklaus/portaudio v0.0.0-20230709114228-aafa478834f5
github.com/k2-fsa/sherpa-onnx-go v1.5.3-alpha.8
github.com/spf13/pflag v1.0.5
)
require (
github.com/k2-fsa/sherpa-onnx-go-linux v1.5.3-alpha.6 // indirect
github.com/k2-fsa/sherpa-onnx-go-macos v1.5.3-alpha.4 // indirect
github.com/k2-fsa/sherpa-onnx-go-windows v1.5.3-alpha.5 // indirect
)
... ...
github.com/gordonklaus/portaudio v0.0.0-20230709114228-aafa478834f5 h1:5AlozfqaVjGYGhms2OsdUyfdJME76E6rx5MdGpjzZpc=
github.com/gordonklaus/portaudio v0.0.0-20230709114228-aafa478834f5/go.mod h1:WY8R6YKlI2ZI3UyzFk7P6yGSuS+hFwNtEzrexRyD7Es=
github.com/k2-fsa/sherpa-onnx-go v1.5.3-alpha.8 h1:BXc31pWwd7CJLM6m9HavxqiyYJdN3Jc9I26pd4x+JHE=
github.com/k2-fsa/sherpa-onnx-go v1.5.3-alpha.8/go.mod h1:kszL/pwg9XTpRGi1AYW/aSwdhRBqb6LN0SjXR0jnsBo=
github.com/k2-fsa/sherpa-onnx-go-linux v1.5.3-alpha.6 h1:o0+l4Wr3IWkWH+kdt8ZZP55L8mRSpW7h1KzvXcfx9FM=
github.com/k2-fsa/sherpa-onnx-go-linux v1.5.3-alpha.6/go.mod h1:9kU02PSdDdzBApwIDBmE2jWS54WvZbRafOEvW/PVLxE=
github.com/k2-fsa/sherpa-onnx-go-macos v1.5.3-alpha.4 h1:Xri23R3+tQFkNwO6zVxqZbjMRJP40z7JtoSqW6UP6sM=
github.com/k2-fsa/sherpa-onnx-go-macos v1.5.3-alpha.4/go.mod h1:a+AJZKNQkFO+JyzGkHySysYfBzzdcoJI5ITFsnhVcmo=
github.com/k2-fsa/sherpa-onnx-go-windows v1.5.3-alpha.5 h1:OR1LLoptR8W35j5u06KrYUblG35B83HHCPWrVK31uHY=
github.com/k2-fsa/sherpa-onnx-go-windows v1.5.3-alpha.5/go.mod h1:ermMOETZUv0nM7MmnXWqeHREMR6zQXBhJpEP2fbHIZo=
github.com/spf13/pflag v1.0.5 h1:iy+VFUOCP1a+8yFto/drg2CJ5u0yRoB7fZw3DKv/JXA=
github.com/spf13/pflag v1.0.5/go.mod h1:McXfInJRrz4CZXVZOBLb0bTZqETkiAhM9Iw0y3An2Bg=
... ...
package main
import (
"github.com/gordonklaus/portaudio"
sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
flag "github.com/spf13/pflag"
"strings"
"fmt"
"log"
)
func main() {
err := portaudio.Initialize()
if err != nil {
log.Fatalf("Unable to initialize portaudio: %v\n", err)
}
defer portaudio.Terminate()
default_device, err := portaudio.DefaultInputDevice()
if err != nil {
log.Fatal("Failed to get default input device: %v\n", err)
}
fmt.Printf("Select default input device: %s\n", default_device.Name)
param := portaudio.StreamParameters{}
param.Input.Device = default_device
param.Input.Channels = 1
param.Input.Latency = default_device.DefaultLowInputLatency
param.SampleRate = 16000
param.FramesPerBuffer = 0
param.Flags = portaudio.ClipOff
config := sherpa.OnlineRecognizerConfig{}
config.FeatConfig = sherpa.FeatureConfig{SampleRate: 16000, FeatureDim: 80}
flag.StringVar(&config.ModelConfig.Encoder, "encoder", "", "Path to the encoder model")
flag.StringVar(&config.ModelConfig.Decoder, "decoder", "", "Path to the decoder model")
flag.StringVar(&config.ModelConfig.Joiner, "joiner", "", "Path to the joiner model")
flag.StringVar(&config.ModelConfig.Tokens, "tokens", "", "Path to the tokens file")
flag.IntVar(&config.ModelConfig.NumThreads, "num-threads", 1, "Number of threads for computing")
flag.IntVar(&config.ModelConfig.Debug, "debug", 0, "Whether to show debug message")
flag.StringVar(&config.ModelConfig.ModelType, "model-type", "", "Optional. Used for loading the model in a faster way")
flag.StringVar(&config.ModelConfig.Provider, "provider", "cpu", "Provider to use")
flag.StringVar(&config.DecodingMethod, "decoding-method", "greedy_search", "Decoding method. Possible values: greedy_search, modified_beam_search")
flag.IntVar(&config.MaxActivePaths, "max-active-paths", 4, "Used only when --decoding-method is modified_beam_search")
flag.IntVar(&config.EnableEndpoint, "enable-endpoint", 1, "Whether to enable endpoint")
flag.Float32Var(&config.Rule1MinTrailingSilence, "rule1-min-trailing-silence", 2.4, "Threshold for rule1")
flag.Float32Var(&config.Rule2MinTrailingSilence, "rule2-min-trailing-silence", 1.2, "Threshold for rule2")
flag.Float32Var(&config.Rule3MinUtteranceLength, "rule3-min-utterance-length", 20, "Threshold for rule3")
flag.Parse()
log.Println("Initializing recognizer (may take several seconds)")
recognizer := sherpa.NewOnlineRecognizer(&config)
log.Println("Recognizer created!")
defer sherpa.DeleteOnlineRecognizer(recognizer)
stream := sherpa.NewOnlineStream(recognizer)
// you can choose another value for 0.1 if you want
samplesPerCall := int32(param.SampleRate * 0.1) // 0.1 second
samples := make([]float32, samplesPerCall)
s, err := portaudio.OpenStream(param, samples)
if err != nil {
log.Fatalf("Failed to open the stream")
}
defer s.Close()
chk(s.Start())
var last_text string
segment_idx := 0
fmt.Println("Started! Please speak")
for {
chk(s.Read())
stream.AcceptWaveform(int(param.SampleRate), samples)
for recognizer.IsReady(stream) {
recognizer.Decode(stream)
}
text := recognizer.GetResult(stream).Text
if len(text) != 0 && last_text != text {
last_text = strings.ToLower(text)
fmt.Printf("\r%d: %s", segment_idx, last_text)
}
if recognizer.IsEndpoint(stream) {
if len(text) != 0 {
segment_idx++
fmt.Println()
}
recognizer.Reset(stream)
}
}
chk(s.Stop())
return
}
func chk(err error) {
if err != nil {
panic(err)
}
}
... ...
#!/usr/bin/env bash
# Please refer to
# https://k2-fsa.github.io/sherpa/onnx/pretrained_models/online-transducer/zipformer-transducer-models.html#pkufool-icefall-asr-zipformer-streaming-wenetspeech-20230615-chinese
# to download the model
# before you run this script.
#
# You can switch to different online models if you need
./real-time-speech-recognition-from-microphone \
--encoder ./icefall-asr-zipformer-streaming-wenetspeech-20230615/exp/encoder-epoch-12-avg-4-chunk-16-left-128.onnx \
--decoder ./icefall-asr-zipformer-streaming-wenetspeech-20230615/exp/decoder-epoch-12-avg-4-chunk-16-left-128.onnx \
--joiner ./icefall-asr-zipformer-streaming-wenetspeech-20230615/exp/joiner-epoch-12-avg-4-chunk-16-left-128.onnx \
--tokens ./icefall-asr-zipformer-streaming-wenetspeech-20230615/data/lang_char/tokens.txt \
--model-type zipformer2
... ...
module streaming-decode-files
go 1.20
require (
github.com/k2-fsa/sherpa-onnx-go v1.5.3-alpha.8
github.com/spf13/pflag v1.0.5
github.com/youpy/go-wav v0.3.2
)
require (
github.com/k2-fsa/sherpa-onnx-go-linux v1.5.3-alpha.6 // indirect
github.com/k2-fsa/sherpa-onnx-go-macos v1.5.3-alpha.4 // indirect
github.com/k2-fsa/sherpa-onnx-go-windows v1.5.3-alpha.5 // indirect
github.com/youpy/go-riff v0.1.0 // indirect
github.com/zaf/g711 v0.0.0-20190814101024-76a4a538f52b // indirect
)
... ...
github.com/davecgh/go-spew v1.1.0 h1:ZDRjVQ15GmhC3fiQ8ni8+OwkZQO4DARzQgrnXU1Liz8=
github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
github.com/google/go-cmp v0.5.6 h1:BKbKCqvP6I+rmFHt06ZmyQtvB8xAkWdhFyr0ZUNZcxQ=
github.com/google/go-cmp v0.5.6/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE=
github.com/k2-fsa/sherpa-onnx-go v1.5.3-alpha.8 h1:BXc31pWwd7CJLM6m9HavxqiyYJdN3Jc9I26pd4x+JHE=
github.com/k2-fsa/sherpa-onnx-go v1.5.3-alpha.8/go.mod h1:kszL/pwg9XTpRGi1AYW/aSwdhRBqb6LN0SjXR0jnsBo=
github.com/k2-fsa/sherpa-onnx-go-linux v1.5.3-alpha.6 h1:o0+l4Wr3IWkWH+kdt8ZZP55L8mRSpW7h1KzvXcfx9FM=
github.com/k2-fsa/sherpa-onnx-go-linux v1.5.3-alpha.6/go.mod h1:9kU02PSdDdzBApwIDBmE2jWS54WvZbRafOEvW/PVLxE=
github.com/k2-fsa/sherpa-onnx-go-macos v1.5.3-alpha.4 h1:Xri23R3+tQFkNwO6zVxqZbjMRJP40z7JtoSqW6UP6sM=
github.com/k2-fsa/sherpa-onnx-go-macos v1.5.3-alpha.4/go.mod h1:a+AJZKNQkFO+JyzGkHySysYfBzzdcoJI5ITFsnhVcmo=
github.com/k2-fsa/sherpa-onnx-go-windows v1.5.3-alpha.5 h1:OR1LLoptR8W35j5u06KrYUblG35B83HHCPWrVK31uHY=
github.com/k2-fsa/sherpa-onnx-go-windows v1.5.3-alpha.5/go.mod h1:ermMOETZUv0nM7MmnXWqeHREMR6zQXBhJpEP2fbHIZo=
github.com/pkg/errors v0.9.1 h1:FEBLx1zS214owpjy7qsBeixbURkuhQAwrK5UwLGTwt4=
github.com/pkg/errors v0.9.1/go.mod h1:bwawxfHBFNV+L2hUp1rHADufV3IMtnDRdf1r5NINEl0=
github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=
github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
github.com/spf13/pflag v1.0.5 h1:iy+VFUOCP1a+8yFto/drg2CJ5u0yRoB7fZw3DKv/JXA=
github.com/spf13/pflag v1.0.5/go.mod h1:McXfInJRrz4CZXVZOBLb0bTZqETkiAhM9Iw0y3An2Bg=
github.com/stretchr/objx v0.1.0/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+wExME=
github.com/stretchr/testify v1.6.1 h1:hDPOHmpOpP40lSULcqw7IrRb/u7w6RpDC9399XyoNd0=
github.com/stretchr/testify v1.6.1/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/h/Wwjteg=
github.com/youpy/go-riff v0.1.0 h1:vZO/37nI4tIET8tQI0Qn0Y79qQh99aEpponTPiPut7k=
github.com/youpy/go-riff v0.1.0/go.mod h1:83nxdDV4Z9RzrTut9losK7ve4hUnxUR8ASSz4BsKXwQ=
github.com/youpy/go-wav v0.3.2 h1:NLM8L/7yZ0Bntadw/0h95OyUsen+DQIVf9gay+SUsMU=
github.com/youpy/go-wav v0.3.2/go.mod h1:0FCieAXAeSdcxFfwLpRuEo0PFmAoc+8NU34h7TUvk50=
github.com/zaf/g711 v0.0.0-20190814101024-76a4a538f52b h1:QqixIpc5WFIqTLxB3Hq8qs0qImAgBdq0p6rq2Qdl634=
github.com/zaf/g711 v0.0.0-20190814101024-76a4a538f52b/go.mod h1:T2h1zV50R/q0CVYnsQOQ6L7P4a2ZxH47ixWcMXFGyx8=
golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c h1:dUUwHk2QECo/6vqA44rthZ8ie2QXMNeKRTHCNY2nXvo=
gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
gotest.tools v2.2.0+incompatible h1:VsBPFP1AI068pPrMxtb/S8Zkgf9xEmTLJjfM+P5UIEo=
gotest.tools v2.2.0+incompatible/go.mod h1:DsYFclhRJ6vuDpmuTbkuFWG+y2sxOXAzmJt81HFBacw=
... ...
package main
import (
"bytes"
"encoding/binary"
sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
flag "github.com/spf13/pflag"
"github.com/youpy/go-wav"
"os"
"strings"
"log"
)
func main() {
log.SetFlags(log.LstdFlags | log.Lmicroseconds)
config := sherpa.OnlineRecognizerConfig{}
config.FeatConfig = sherpa.FeatureConfig{SampleRate: 16000, FeatureDim: 80}
flag.StringVar(&config.ModelConfig.Encoder, "encoder", "", "Path to the encoder model")
flag.StringVar(&config.ModelConfig.Decoder, "decoder", "", "Path to the decoder model")
flag.StringVar(&config.ModelConfig.Joiner, "joiner", "", "Path to the joiner model")
flag.StringVar(&config.ModelConfig.Tokens, "tokens", "", "Path to the tokens file")
flag.IntVar(&config.ModelConfig.NumThreads, "num-threads", 1, "Number of threads for computing")
flag.IntVar(&config.ModelConfig.Debug, "debug", 0, "Whether to show debug message")
flag.StringVar(&config.ModelConfig.ModelType, "model-type", "", "Optional. Used for loading the model in a faster way")
flag.StringVar(&config.ModelConfig.Provider, "provider", "cpu", "Provider to use")
flag.StringVar(&config.DecodingMethod, "decoding-method", "greedy_search", "Decoding method. Possible values: greedy_search, modified_beam_search")
flag.IntVar(&config.MaxActivePaths, "max-active-paths", 4, "Used only when --decoding-method is modified_beam_search")
flag.Parse()
if len(flag.Args()) != 1 {
log.Fatalf("Please provide one wave file")
}
log.Println("Reading", flag.Arg(0))
samples, sampleRate := readWave(flag.Arg(0))
log.Println("Initializing recognizer (may take several seconds)")
recognizer := sherpa.NewOnlineRecognizer(&config)
log.Println("Recognizer created!")
defer sherpa.DeleteOnlineRecognizer(recognizer)
log.Println("Start decoding!")
stream := sherpa.NewOnlineStream(recognizer)
defer sherpa.DeleteOnlineStream(stream)
stream.AcceptWaveform(sampleRate, samples)
tailPadding := make([]float32, int(float32(sampleRate)*0.3))
stream.AcceptWaveform(sampleRate, tailPadding)
for recognizer.IsReady(stream) {
recognizer.Decode(stream)
}
log.Println("Decoding done!")
result := recognizer.GetResult(stream)
log.Println(strings.ToLower(result.Text))
log.Printf("Wave duration: %v seconds", float32(len(samples))/float32(sampleRate))
}
func readWave(filename string) (samples []float32, sampleRate int) {
file, _ := os.Open(filename)
defer file.Close()
reader := wav.NewReader(file)
format, err := reader.Format()
if err != nil {
log.Fatalf("Failed to read wave format")
}
if format.AudioFormat != 1 {
log.Fatalf("Support only PCM format. Given: %v\n", format.AudioFormat)
}
if format.NumChannels != 1 {
log.Fatalf("Support only 1 channel wave file. Given: %v\n", format.NumChannels)
}
if format.BitsPerSample != 16 {
log.Fatalf("Support only 16-bit per sample. Given: %v\n", format.BitsPerSample)
}
reader.Duration() // so that it initializes reader.Size
buf := make([]byte, reader.Size)
n, err := reader.Read(buf)
if n != int(reader.Size) {
log.Fatalf("Failed to read %v bytes. Returned %v bytes\n", reader.Size, n)
}
samples = samplesInt16ToFloat(buf)
sampleRate = int(format.SampleRate)
return
}
func samplesInt16ToFloat(inSamples []byte) []float32 {
numSamples := len(inSamples) / 2
outSamples := make([]float32, numSamples)
for i := 0; i != numSamples; i++ {
s := inSamples[i*2 : (i+1)*2]
var s16 int16
buf := bytes.NewReader(s)
err := binary.Read(buf, binary.LittleEndian, &s16)
if err != nil {
log.Fatal("Failed to parse 16-bit sample")
}
outSamples[i] = float32(s16) / 32768
}
return outSamples
}
... ...
#!/usr/bin/env bash
# Please refer to
# https://k2-fsa.github.io/sherpa/onnx/pretrained_models/online-transducer/zipformer-transducer-models.html#csukuangfj-sherpa-onnx-streaming-zipformer-en-2023-06-26-english
# to download the model
# before you run this script.
#
# You can switch to a different online model if you need
./streaming-decode-files \
--encoder ./sherpa-onnx-streaming-zipformer-en-2023-06-26/encoder-epoch-99-avg-1-chunk-16-left-128.onnx \
--decoder ./sherpa-onnx-streaming-zipformer-en-2023-06-26/decoder-epoch-99-avg-1-chunk-16-left-128.onnx \
--joiner ./sherpa-onnx-streaming-zipformer-en-2023-06-26/joiner-epoch-99-avg-1-chunk-16-left-128.onnx \
--tokens ./sherpa-onnx-streaming-zipformer-en-2023-06-26/tokens.txt \
--model-type zipformer2 \
--debug 0 \
./sherpa-onnx-streaming-zipformer-en-2023-06-26/test_wavs/0.wav
... ...
#!/usr/bin/env bash
set -ex
git config --global user.email "csukuangfj@gmail.com"
git config --global user.name "Fangjun Kuang"
SHERPA_ONNX_VERSION=v$(grep "SHERPA_ONNX_VERSION" ./CMakeLists.txt | cut -d " " -f 2 | cut -d '"' -f 2)
echo "========================================================================="
git clone git@github.com:k2-fsa/sherpa-onnx-go-linux.git
echo "Copy libs for Linux x86_64"
rm -rf sherpa-onnx-go-linux/lib/x86_64-unknown-linux-gnu/lib*
cp -v ./linux/sherpa_onnx/lib/libkaldi-native-fbank-core.so sherpa-onnx-go-linux/lib/x86_64-unknown-linux-gnu/
cp -v ./linux/sherpa_onnx/lib/libonnxruntime* sherpa-onnx-go-linux/lib/x86_64-unknown-linux-gnu/
cp -v ./linux/sherpa_onnx/lib/libsherpa-onnx-c-api.so sherpa-onnx-go-linux/lib/x86_64-unknown-linux-gnu/
cp -v ./linux/sherpa_onnx/lib/libsherpa-onnx-core.so sherpa-onnx-go-linux/lib/x86_64-unknown-linux-gnu/
echo "Copy sources for Linux x86_64"
cp sherpa-onnx/c-api/c-api.h sherpa-onnx-go-linux/
cp scripts/go/sherpa_onnx.go sherpa-onnx-go-linux/
pushd sherpa-onnx-go-linux
tag=$(git describe --abbrev=0 --tags)
if [[ x"$VERSION" == x"auto" ]]; then
# this is a pre-release
if [[ $tag == ${SHERPA_ONNX_VERSION}* ]]; then
# echo we have already release pre-release before, so just increment it
last=$(echo $tag | rev | cut -d'.' -f 1 | rev)
new_last=$((last+1))
new_tag=${SHERPA_ONNX_VERSION}-alpha.${new_last}
else
new_tag=${SHERPA_ONNX_VERSION}-alpha.1
fi
else
new_tag=$VERSION
fi
echo "new_tag: $new_tag"
git add .
git status
git commit -m "Release $new_tag" && \
git tag $new_tag && \
git push origin $new_tag || true
popd
echo "========================================================================="
git clone git@github.com:k2-fsa/sherpa-onnx-go-macos.git
echo "Copy libs for macOS x86_64"
rm -rf sherpa-onnx-go-macos/lib/x86_64-apple-darwin/lib*
cp -v ./macos-x86_64/libkaldi-native-fbank-core.dylib sherpa-onnx-go-macos/lib/x86_64-apple-darwin
cp -v ./macos-x86_64/libonnxruntime* sherpa-onnx-go-macos/lib/x86_64-apple-darwin
cp -v ./macos-x86_64/libsherpa-onnx-c-api.dylib sherpa-onnx-go-macos/lib/x86_64-apple-darwin
cp -v ./macos-x86_64/libsherpa-onnx-core.dylib sherpa-onnx-go-macos/lib/x86_64-apple-darwin
echo "Copy libs for macOS arm64"
rm -rf sherpa-onnx-go-macos/lib/aarch64-apple-darwin/lib*
cp -v ./macos-arm64/libkaldi-native-fbank-core.dylib sherpa-onnx-go-macos/lib/aarch64-apple-darwin
cp -v ./macos-arm64/libonnxruntime* sherpa-onnx-go-macos/lib/aarch64-apple-darwin
cp -v ./macos-arm64/libsherpa-onnx-c-api.dylib sherpa-onnx-go-macos/lib/aarch64-apple-darwin
cp -v ./macos-arm64/libsherpa-onnx-core.dylib sherpa-onnx-go-macos/lib/aarch64-apple-darwin
echo "Copy sources for macOS"
cp sherpa-onnx/c-api/c-api.h sherpa-onnx-go-macos/
cp scripts/go/sherpa_onnx.go sherpa-onnx-go-macos/
pushd sherpa-onnx-go-macos
tag=$(git describe --abbrev=0 --tags)
if [[ x"$VERSION" == x"auto" ]]; then
# this is a pre-release
if [[ $tag == ${SHERPA_ONNX_VERSION}* ]]; then
# echo we have already release pre-release before, so just increment it
last=$(echo $tag | rev | cut -d'.' -f 1 | rev)
new_last=$((last+1))
new_tag=${SHERPA_ONNX_VERSION}-alpha.${new_last}
else
new_tag=${SHERPA_ONNX_VERSION}-alpha.1
fi
else
new_tag=$VERSION
fi
echo "new_tag: $new_tag"
git add .
git status
git commit -m "Release $new_tag" && \
git tag $new_tag && \
git push origin $new_tag || true
popd
echo "========================================================================="
git clone git@github.com:k2-fsa/sherpa-onnx-go-windows.git
echo "Copy libs for Windows x86_64"
rm -fv sherpa-onnx-go-windows/lib/x86_64-pc-windows-gnu/*
cp -v ./windows-x64/kaldi-native-fbank-core.dll sherpa-onnx-go-windows/lib/x86_64-pc-windows-gnu
cp -v ./windows-x64/onnxruntime.dll sherpa-onnx-go-windows/lib/x86_64-pc-windows-gnu
cp -v ./windows-x64/sherpa-onnx-c-api.dll sherpa-onnx-go-windows/lib/x86_64-pc-windows-gnu
cp -v ./windows-x64/sherpa-onnx-core.dll sherpa-onnx-go-windows/lib/x86_64-pc-windows-gnu
echo "Copy libs for Windows x86"
rm -fv sherpa-onnx-go-windows/lib/i686-pc-windows-gnu/*
cp -v ./windows-win32/kaldi-native-fbank-core.dll sherpa-onnx-go-windows/lib/i686-pc-windows-gnu
cp -v ./windows-win32/onnxruntime.dll sherpa-onnx-go-windows/lib/i686-pc-windows-gnu
cp -v ./windows-win32/sherpa-onnx-c-api.dll sherpa-onnx-go-windows/lib/i686-pc-windows-gnu
cp -v ./windows-win32/sherpa-onnx-core.dll sherpa-onnx-go-windows/lib/i686-pc-windows-gnu
echo "Copy sources for Windows"
cp sherpa-onnx/c-api/c-api.h sherpa-onnx-go-windows/
cp scripts/go/sherpa_onnx.go sherpa-onnx-go-windows/
pushd sherpa-onnx-go-windows
tag=$(git describe --abbrev=0 --tags)
if [[ x"$VERSION" == x"auto" ]]; then
# this is a pre-release
if [[ $tag == ${SHERPA_ONNX_VERSION}* ]]; then
# echo we have already release pre-release before, so just increment it
last=$(echo $tag | rev | cut -d'.' -f 1 | rev)
new_last=$((last+1))
new_tag=${SHERPA_ONNX_VERSION}-alpha.${new_last}
else
new_tag=${SHERPA_ONNX_VERSION}-alpha.1
fi
else
new_tag=$VERSION
fi
echo "new_tag: $new_tag"
git add .
git status
git commit -m "Release $new_tag" && \
git tag $new_tag && \
git push origin $new_tag || true
popd
echo "========================================================================="
rm -fv ~/.ssh/github
... ...
/*
Speech recognition with [Next-gen Kaldi].
[sherpa-onnx] is an open-source speech recognition framework for [Next-gen Kaldi].
It depends only on [onnxruntime], supporting both streaming and non-streaming
speech recognition.
It does not need to access the network during recognition and everything
runs locally.
It supports a variety of platforms, such as Linux (x86_64, aarch64, arm),
Windows (x86_64, x86), macOS (x86_64, arm64), etc.
Usage examples:
1. Real-time speech recognition from a microphone
Please see
https://github.com/k2-fsa/sherpa-onnx/tree/master/go-api-examples/real-time-speech-recognition-from-microphone
2. Decode files using a non-streaming model
Please see
https://github.com/k2-fsa/sherpa-onnx/tree/master/go-api-examples/non-streaming-decode-files
3. Decode files using a streaming model
Please see
https://github.com/k2-fsa/sherpa-onnx/tree/master/go-api-examples/streaming-decode-files
[sherpa-onnx]: https://github.com/k2-fsa/sherpa-onnx
[onnxruntime]: https://github.com/microsoft/onnxruntime
[Next-gen Kaldi]: https://github.com/k2-fsa/
*/
package sherpa_onnx
// #include <stdlib.h>
// #include "c-api.h"
import "C"
import "unsafe"
// Configuration for online/streaming transducer models
//
// Please refer to
// https://k2-fsa.github.io/sherpa/onnx/pretrained_models/online-transducer/index.html
// to download pre-trained models
type OnlineTransducerModelConfig struct {
Encoder string // Path to the encoder model, e.g., encoder.onnx or encoder.int8.onnx
Decoder string // Path to the decoder model.
Joiner string // Path to the joiner model.
Tokens string // Path to tokens.txt
NumThreads int // Number of threads to use for neural network computation
Provider string // Optional. Valid values are: cpu, cuda, coreml
Debug int // 1 to show model meta information while loading it.
ModelType string // Optional. You can specify it for faster model initialization
}
// Configuration for the feature extractor
type FeatureConfig struct {
// Sample rate expected by the model. It is 16000 for all
// pre-trained models provided by us
SampleRate int
// Feature dimension expected by the model. It is 80 for all
// pre-trained models provided by us
FeatureDim int
}
// Configuration for the online/streaming recognizer.
type OnlineRecognizerConfig struct {
FeatConfig FeatureConfig
ModelConfig OnlineTransducerModelConfig
// Valid decoding methods: greedy_search, modified_beam_search
DecodingMethod string
// Used only when DecodingMethod is modified_beam_search. It specifies
// the maximum number of paths to keep during the search
MaxActivePaths int
EnableEndpoint int // 1 to enable endpoint detection.
// Please see
// https://k2-fsa.github.io/sherpa/ncnn/endpoint.html
// for the meaning of Rule1MinTrailingSilence, Rule2MinTrailingSilence
// and Rule3MinUtteranceLength.
Rule1MinTrailingSilence float32
Rule2MinTrailingSilence float32
Rule3MinUtteranceLength float32
}
// It contains the recognition result for a online stream.
type OnlineRecognizerResult struct {
Text string
}
// The online recognizer class. It wraps a pointer from C.
type OnlineRecognizer struct {
impl *C.struct_SherpaOnnxOnlineRecognizer
}
// The online stream class. It wraps a pointer from C.
type OnlineStream struct {
impl *C.struct_SherpaOnnxOnlineStream
}
// Free the internal pointer inside the recognizer to avoid memory leak.
func DeleteOnlineRecognizer(recognizer *OnlineRecognizer) {
C.DestroyOnlineRecognizer(recognizer.impl)
recognizer.impl = nil
}
// The user is responsible to invoke [DeleteOnlineRecognizer]() to free
// the returned recognizer to avoid memory leak
func NewOnlineRecognizer(config *OnlineRecognizerConfig) *OnlineRecognizer {
c := C.struct_SherpaOnnxOnlineRecognizerConfig{}
c.feat_config.sample_rate = C.int(config.FeatConfig.SampleRate)
c.feat_config.feature_dim = C.int(config.FeatConfig.FeatureDim)
c.model_config.encoder = C.CString(config.ModelConfig.Encoder)
defer C.free(unsafe.Pointer(c.model_config.encoder))
c.model_config.decoder = C.CString(config.ModelConfig.Decoder)
defer C.free(unsafe.Pointer(c.model_config.decoder))
c.model_config.joiner = C.CString(config.ModelConfig.Joiner)
defer C.free(unsafe.Pointer(c.model_config.joiner))
c.model_config.tokens = C.CString(config.ModelConfig.Tokens)
defer C.free(unsafe.Pointer(c.model_config.tokens))
c.model_config.num_threads = C.int(config.ModelConfig.NumThreads)
c.model_config.provider = C.CString(config.ModelConfig.Provider)
defer C.free(unsafe.Pointer(c.model_config.provider))
c.model_config.debug = C.int(config.ModelConfig.Debug)
c.model_config.model_type = C.CString(config.ModelConfig.ModelType)
defer C.free(unsafe.Pointer(c.model_config.model_type))
c.decoding_method = C.CString(config.DecodingMethod)
defer C.free(unsafe.Pointer(c.decoding_method))
c.max_active_paths = C.int(config.MaxActivePaths)
c.enable_endpoint = C.int(config.EnableEndpoint)
c.rule1_min_trailing_silence = C.float(config.Rule1MinTrailingSilence)
c.rule2_min_trailing_silence = C.float(config.Rule2MinTrailingSilence)
c.rule3_min_utterance_length = C.float(config.Rule3MinUtteranceLength)
recognizer := &OnlineRecognizer{}
recognizer.impl = C.CreateOnlineRecognizer(&c)
return recognizer
}
// Delete the internal pointer inside the stream to avoid memory leak.
func DeleteOnlineStream(stream *OnlineStream) {
C.DestroyOnlineStream(stream.impl)
stream.impl = nil
}
// The user is responsible to invoke [DeleteOnlineStream]() to free
// the returned stream to avoid memory leak
func NewOnlineStream(recognizer *OnlineRecognizer) *OnlineStream {
stream := &OnlineStream{}
stream.impl = C.CreateOnlineStream(recognizer.impl)
return stream
}
// Input audio samples for the stream.
//
// sampleRate is the actual sample rate of the input audio samples. If it
// is different from the sample rate expected by the feature extractor, we will
// do resampling inside.
//
// samples contains audio samples. Each sample is in the range [-1, 1]
func (s *OnlineStream) AcceptWaveform(sampleRate int, samples []float32) {
C.AcceptWaveform(s.impl, C.int(sampleRate), (*C.float)(&samples[0]), C.int(len(samples)))
}
// Signal that there will be no incoming audio samples.
// After calling this function, you cannot call [OnlineStream.AcceptWaveform] any longer.
//
// The main purpose of this function is to flush the remaining audio samples
// buffered inside for feature extraction.
func (s *OnlineStream) InputFinished() {
C.InputFinished(s.impl)
}
// Check whether the stream has enough feature frames for decoding.
// Return true if this stream is ready for decoding. Return false otherwise.
//
// You will usually use it like below:
//
// for recognizer.IsReady(s) {
// recognizer.Decode(s)
// }
func (recognizer *OnlineRecognizer) IsReady(s *OnlineStream) bool {
return C.IsOnlineStreamReady(recognizer.impl, s.impl) == 1
}
// Return true if an endpoint is detected.
//
// You usually use it like below:
//
// if recognizer.IsEndpoint(s) {
// // do your own stuff after detecting an endpoint
//
// recognizer.Reset(s)
// }
func (recognizer *OnlineRecognizer) IsEndpoint(s *OnlineStream) bool {
return C.IsEndpoint(recognizer.impl, s.impl) == 1
}
// After calling this function, the internal neural network model states
// are reset and IsEndpoint(s) would return false. GetResult(s) would also
// return an empty string.
func (recognizer *OnlineRecognizer) Reset(s *OnlineStream) {
C.Reset(recognizer.impl, s.impl)
}
// Decode the stream. Before calling this function, you have to ensure
// that recognizer.IsReady(s) returns true. Otherwise, you will be SAD.
//
// You usually use it like below:
//
// for recognizer.IsReady(s) {
// recognizer.Decode(s)
// }
func (recognizer *OnlineRecognizer) Decode(s *OnlineStream) {
C.DecodeOnlineStream(recognizer.impl, s.impl)
}
// Decode multiple streams in parallel, i.e., in batch.
// You have to ensure that each stream is ready for decoding. Otherwise,
// you will be SAD.
func (recognizer *OnlineRecognizer) DecodeStreams(s []*OnlineStream) {
ss := make([]*C.struct_SherpaOnnxOnlineStream, len(s))
for i, v := range s {
ss[i] = v.impl
}
C.DecodeMultipleOnlineStreams(recognizer.impl, &ss[0], C.int(len(s)))
}
// Get the current result of stream since the last invoke of Reset()
func (recognizer *OnlineRecognizer) GetResult(s *OnlineStream) *OnlineRecognizerResult {
p := C.GetOnlineStreamResult(recognizer.impl, s.impl)
defer C.DestroyOnlineRecognizerResult(p)
result := &OnlineRecognizerResult{}
result.Text = C.GoString(p.text)
return result
}
// Configuration for offline/non-streaming transducer.
//
// Please refer to
// https://k2-fsa.github.io/sherpa/onnx/pretrained_models/offline-transducer/index.html
// to download pre-trained models
type OfflineTransducerModelConfig struct {
Encoder string // Path to the encoder model, i.e., encoder.onnx or encoder.int8.onnx
Decoder string // Path to the decoder model
Joiner string // Path to the joiner model
}
// Configuration for offline/non-streaming paraformer.
//
// please refer to
// https://k2-fsa.github.io/sherpa/onnx/pretrained_models/offline-paraformer/index.html
// to download pre-trained models
type OfflineParaformerModelConfig struct {
Model string // Path to the model, e.g., model.onnx or model.int8.onnx
}
// Configuration for offline/non-streaming NeMo CTC models.
//
// Please refer to
// https://k2-fsa.github.io/sherpa/onnx/pretrained_models/offline-ctc/index.html
// to download pre-trained models
type OfflineNemoEncDecCtcModelConfig struct {
Model string // Path to the model, e.g., model.onnx or model.int8.onnx
}
// Configuration for offline LM.
type OfflineLMConfig struct {
Model string // Path to the model
Scale float32 // scale for LM score
}
type OfflineModelConfig struct {
Transducer OfflineTransducerModelConfig
Paraformer OfflineParaformerModelConfig
NemoCTC OfflineNemoEncDecCtcModelConfig
Tokens string // Path to tokens.txt
// Number of threads to use for neural network computation
NumThreads int
// 1 to print model meta information while loading
Debug int
// Optional. Valid values: cpu, cuda, coreml
Provider string
// Optional. Specify it for faster model initialization.
ModelType string
}
// Configuration for the offline/non-streaming recognizer.
type OfflineRecognizerConfig struct {
FeatConfig FeatureConfig
ModelConfig OfflineModelConfig
LmConfig OfflineLMConfig
// Valid decoding method: greedy_search, modified_beam_search
DecodingMethod string
// Used only when DecodingMethod is modified_beam_search.
MaxActivePaths int
}
// It wraps a pointer from C
type OfflineRecognizer struct {
impl *C.struct_SherpaOnnxOfflineRecognizer
}
// It wraps a pointer from C
type OfflineStream struct {
impl *C.struct_SherpaOnnxOfflineStream
}
// It contains recognition result of an offline stream.
type OfflineRecognizerResult struct {
Text string
}
// Frees the internal pointer of the recognition to avoid memory leak.
func DeleteOfflineRecognizer(recognizer *OfflineRecognizer) {
C.DestroyOfflineRecognizer(recognizer.impl)
recognizer.impl = nil
}
// The user is responsible to invoke [DeleteOfflineRecognizer]() to free
// the returned recognizer to avoid memory leak
func NewOfflineRecognizer(config *OfflineRecognizerConfig) *OfflineRecognizer {
c := C.struct_SherpaOnnxOfflineRecognizerConfig{}
c.feat_config.sample_rate = C.int(config.FeatConfig.SampleRate)
c.feat_config.feature_dim = C.int(config.FeatConfig.FeatureDim)
c.model_config.transducer.encoder = C.CString(config.ModelConfig.Transducer.Encoder)
defer C.free(unsafe.Pointer(c.model_config.transducer.encoder))
c.model_config.transducer.decoder = C.CString(config.ModelConfig.Transducer.Decoder)
defer C.free(unsafe.Pointer(c.model_config.transducer.decoder))
c.model_config.transducer.joiner = C.CString(config.ModelConfig.Transducer.Joiner)
defer C.free(unsafe.Pointer(c.model_config.transducer.joiner))
c.model_config.paraformer.model = C.CString(config.ModelConfig.Paraformer.Model)
defer C.free(unsafe.Pointer(c.model_config.paraformer.model))
c.model_config.nemo_ctc.model = C.CString(config.ModelConfig.NemoCTC.Model)
defer C.free(unsafe.Pointer(c.model_config.nemo_ctc.model))
c.model_config.tokens = C.CString(config.ModelConfig.Tokens)
defer C.free(unsafe.Pointer(c.model_config.tokens))
c.model_config.num_threads = C.int(config.ModelConfig.NumThreads)
c.model_config.debug = C.int(config.ModelConfig.Debug)
c.model_config.provider = C.CString(config.ModelConfig.Provider)
defer C.free(unsafe.Pointer(c.model_config.provider))
c.model_config.model_type = C.CString(config.ModelConfig.ModelType)
defer C.free(unsafe.Pointer(c.model_config.model_type))
c.lm_config.model = C.CString(config.LmConfig.Model)
defer C.free(unsafe.Pointer(c.lm_config.model))
c.lm_config.scale = C.float(config.LmConfig.Scale)
c.decoding_method = C.CString(config.DecodingMethod)
defer C.free(unsafe.Pointer(c.decoding_method))
c.max_active_paths = C.int(config.MaxActivePaths)
recognizer := &OfflineRecognizer{}
recognizer.impl = C.CreateOfflineRecognizer(&c)
return recognizer
}
// Frees the internal pointer of the stream to avoid memory leak.
func DeleteOfflineStream(stream *OfflineStream) {
C.DestroyOfflineStream(stream.impl)
stream.impl = nil
}
// The user is responsible to invoke [DeleteOfflineStream]() to free
// the returned stream to avoid memory leak
func NewOfflineStream(recognizer *OfflineRecognizer) *OfflineStream {
stream := &OfflineStream{}
stream.impl = C.CreateOfflineStream(recognizer.impl)
return stream
}
// Input audio samples for the offline stream.
// Please only call it once. That is, input all samples at once.
//
// sampleRate is the sample rate of the input audio samples. If it is different
// from the value expected by the feature extractor, we will do resampling inside.
//
// samples contains the actual audio samples. Each sample is in the range [-1, 1].
func (s *OfflineStream) AcceptWaveform(sampleRate int, samples []float32) {
C.AcceptWaveformOffline(s.impl, C.int(sampleRate), (*C.float)(&samples[0]), C.int(len(samples)))
}
// Decode the offline stream.
func (recognizer *OfflineRecognizer) Decode(s *OfflineStream) {
C.DecodeOfflineStream(recognizer.impl, s.impl)
}
// Decode multiple streams in parallel, i.e., in batch.
func (recognizer *OfflineRecognizer) DecodeStreams(s []*OfflineStream) {
ss := make([]*C.struct_SherpaOnnxOfflineStream, len(s))
for i, v := range s {
ss[i] = v.impl
}
C.DecodeMultipleOfflineStreams(recognizer.impl, &ss[0], C.int(len(s)))
}
// Get the recognition result of the offline stream.
func (s *OfflineStream) GetResult() *OfflineRecognizerResult {
p := C.GetOfflineStreamResult(s.impl)
defer C.DestroyOfflineRecognizerResult(p)
result := &OfflineRecognizerResult{}
result.Text = C.GoString(p.text)
return result
}
... ...
Host github.com
Hostname github.com
User git
IdentityFile ~/.ssh/github
StrictHostKeyChecking no
... ...