xuning / sherpaonnx

转到一个项目
Toggle navigation Toggle navigation pinning
Lovemefan
Committed by GitHub
b1f08c0a 1 parent dd21ebda

scale value fix (#1006)

内嵌 并排对比
正在显示 3 个修改的文件 包含 3 行增加 和 3 行删除
@@ -80,7 +80,7 @@ def get_features(test_wav_filename): @@ -80,7 +80,7 @@ def get_features(test_wav_filename):
80 samples = librosa.resample(samples, orig_sr=sample_rate, target_sr=16000) 80 samples = librosa.resample(samples, orig_sr=sample_rate, target_sr=16000)
81 sample_rate = 16000 81 sample_rate = 16000
82 82
83 - samples *= 372768 83 + samples *= 32768
84 84
85 opts = knf.MfccOptions() 85 opts = knf.MfccOptions()
86 # See https://github.com/Tele-AI/TeleSpeech-ASR/blob/master/mfcc_hires.conf 86 # See https://github.com/Tele-AI/TeleSpeech-ASR/blob/master/mfcc_hires.conf
@@ -112,7 +112,7 @@ def get_features(test_wav_filename): @@ -112,7 +112,7 @@ def get_features(test_wav_filename):
112 audio = torchaudio.functional.resample( 112 audio = torchaudio.functional.resample(
113 audio, orig_freq=sample_rate, new_freq=16000 113 audio, orig_freq=sample_rate, new_freq=16000
114 ) 114 )
115 - audio *= 372768 115 + audio *= 32768
116 116
117 opts = knf.FbankOptions() 117 opts = knf.FbankOptions()
118 opts.frame_opts.dither = 0 118 opts.frame_opts.dither = 0
@@ -52,7 +52,7 @@ def get_features(test_wav_filename): @@ -52,7 +52,7 @@ def get_features(test_wav_filename):
52 audio = torchaudio.functional.resample( 52 audio = torchaudio.functional.resample(
53 audio, orig_freq=sample_rate, new_freq=16000 53 audio, orig_freq=sample_rate, new_freq=16000
54 ) 54 )
55 - audio *= 372768 55 + audio *= 32768
56 56
57 opts = knf.FbankOptions() 57 opts = knf.FbankOptions()
58 opts.frame_opts.dither = 0 58 opts.frame_opts.dither = 0