Add kitten tts nano v0.2 (#2512)

@@ -3,7 +3,7 @@ name: export-kitten-to-onnx
 on:
   push:
     branches:
-      - kitten-tts
+      - kitten-0.2
 
   workflow_dispatch:
 
@@ -20,6 +20,7 @@ jobs:
       fail-fast: false
       matrix:
         os: [ubuntu-latest]
+        version: ["nano_v0_1", "nano_v0_2"]
         python-version: ["3.10"]
 
     steps:
@@ -40,7 +41,7 @@ jobs:
           HF_TOKEN: ${{ secrets.HF_TOKEN }}
         shell: bash
         run: |
-          cd scripts/kitten-tts/nano_v0_1
+          cd scripts/kitten-tts/${{ matrix.version }}
           ./run.sh
 
       - name: Collect results
@@ -50,9 +51,18 @@ jobs:
           tar xf espeak-ng-data.tar.bz2
           rm espeak-ng-data.tar.bz2
 
-          src=scripts/kitten-tts/nano_v0_1
+          version=${{ matrix.version }}
 
+          src=scripts/kitten-tts/$version
+
+          if [[ $version == "nano_v0_1" ]]; then
             d=kitten-nano-en-v0_1-fp16
+          elif [[ $version == "nano_v0_2" ]]; then
+            d=kitten-nano-en-v0_2-fp16
+          else
+            echo "version $version"
+            exit 1
+          fi
 
           mkdir $d
           cp -a LICENSE $d/LICENSE
@@ -100,12 +110,16 @@ jobs:
 
             dirs=(
               kitten-nano-en-v0_1-fp16
+              kitten-nano-en-v0_2-fp16
             )
 
             export GIT_LFS_SKIP_SMUDGE=1
             export GIT_CLONE_PROTECTION_ACTIVE=false
 
             for d in ${dirs[@]}; do
+              if [ ! -d ../$d ]]; then
+                continue
+              fi
               rm -rf huggingface
 
               git clone https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/$d huggingface

@@ -515,7 +515,12 @@ def get_kitten_models() -> List[TtsModel]:
             model_dir="kitten-nano-en-v0_1-fp16",
             model_name="model.fp16.onnx",
             lang="en",
-        )
+        ),
+        TtsModel(
+            model_dir="kitten-nano-en-v0_2-fp16",
+            model_name="model.fp16.onnx",
+            lang="en",
+        ),
     ]
     for m in english_models:
         m.data_dir = f"{m.model_dir}/espeak-ng-data"

+#!/usr/bin/env python3
+# Copyright    2025  Xiaomi Corp.        (authors: Fangjun Kuang)
+
+
+import argparse
+
+import numpy as np
+import onnx
+
+from generate_voices_bin import speaker2id
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--model", type=str, required=True, help="input and output onnx model"
+    )
+
+    return parser.parse_args()
+
+
+def main():
+    args = get_args()
+    print(args.model)
+
+    model = onnx.load(args.model)
+
+    style = np.load("./voices.npz")
+    style_shape = style[list(style.keys())[0]].shape
+
+    speaker2id_str = ""
+    id2speaker_str = ""
+    sep = ""
+    for s, i in speaker2id.items():
+        speaker2id_str += f"{sep}{s}->{i}"
+        id2speaker_str += f"{sep}{i}->{s}"
+        sep = ","
+
+    meta_data = {
+        "model_type": "kitten-tts",
+        "language": "English",
+        "has_espeak": 1,
+        "sample_rate": 24000,
+        "version": 1,
+        "voice": "en-us",
+        "style_dim": ",".join(map(str, style_shape)),
+        "n_speakers": len(speaker2id),
+        "speaker2id": speaker2id_str,
+        "id2speaker": id2speaker_str,
+        "speaker_names": ",".join(map(str, speaker2id.keys())),
+        "model_url": "https://huggingface.co/KittenML/kitten-tts-nano-0.2",
+        "see_also": "https://github.com/KittenML/KittenTTS",
+        "maintainer": "k2-fsa",
+        "comment": "This is kitten-tts-nano-0.2 and supports only English",
+    }
+
+    print(model.metadata_props)
+
+    while len(model.metadata_props):
+        model.metadata_props.pop()
+
+    for key, value in meta_data.items():
+        meta = model.metadata_props.add()
+        meta.key = key
+        meta.value = str(value)
+    print("--------------------")
+
+    print(model.metadata_props)
+
+    onnx.save(model, args.model)
+
+    print(f"Please see {args.model}")
+
+
+if __name__ == "__main__":
+    main()

+#!/usr/bin/env python3
+# Copyright    2025  Xiaomi Corp.        (authors: Fangjun Kuang)
+
+"""
+Change the model so that it can be run in onnxruntime 1.17.1
+"""
+
+import onnx
+
+
+def main():
+    model = onnx.load("kitten_tts_nano_v0_2.onnx")
+
+    # Print current opsets
+    for opset in model.opset_import:
+        print(f"Domain: '{opset.domain}', Version: {opset.version}")
+
+    # Modify the opset versions (be careful!)
+    for opset in model.opset_import:
+        if opset.domain == "":  # ai.onnx domain
+            opset.version = 19  # change from 20 to 19
+        elif opset.domain == "ai.onnx.ml":
+            opset.version = 4  # change from 5 to 4
+
+    # Save the modified model
+    onnx.save(model, "model.fp16.onnx")
+
+
+if __name__ == "__main__":
+    main()

+../nano_v0_1/generate_tokens.py
\ No newline at end of file

+../nano_v0_1/generate_voices_bin.py
\ No newline at end of file

+#!/usr/bin/env bash
+# Copyright      2025  Xiaomi Corp.        (authors: Fangjun Kuang)
+
+set -ex
+
+if [ ! -f kitten_tts_nano_v0_2.onnx ]; then
+  curl -SL -O https://huggingface.co/KittenML/kitten-tts-nano-0.2/resolve/main/kitten_tts_nano_v0_2.onnx
+fi
+
+if [ ! -f voices.npz ]; then
+  curl -SL -O https://huggingface.co/KittenML/kitten-tts-nano-0.2/resolve/main/voices.npz
+fi
+
+./generate_voices_bin.py
+./generate_tokens.py
+./convert_opset.py
+./show.py
+./add_meta_data.py --model ./model.fp16.onnx
+# ./test.py --model ./model.fp16.onnx --tokens ./tokens.txt --voice ./voices.bin
+ls -lh

+../nano_v0_1/show.py
\ No newline at end of file

+../nano_v0_1/test.py
\ No newline at end of file