mirror of
https://github.com/k2-fsa/sherpa-onnx.git
synced 2026-11-10 21:11:03 +08:00
107 lines
3.6 KiB
Python
107 lines
3.6 KiB
Python
#!/usr/bin/env python3
|
|
|
|
"""
|
|
This file shows how to use the online speech enhancement API with VoxLock.
|
|
|
|
VoxLock is a target-speaker voice isolation model. It keeps the voice of one
|
|
speaker, e.g., the caller in a phone call, and removes the other voices.
|
|
|
|
The model is from https://github.com/arbazsiddiqui/voxlock
|
|
|
|
This script does not provide a voice sample, so VoxLock passes the audio
|
|
through unchanged until it has heard 3 seconds of speech, takes that speech
|
|
as the speaker to keep, and isolates from then on.
|
|
|
|
If you have a voice sample of the speaker, please see
|
|
./online-speech-enhancement-voxlock-with-voice-sample.py instead.
|
|
|
|
Please download files used in this script from
|
|
https://github.com/k2-fsa/sherpa-onnx/releases/tag/speech-enhancement-models
|
|
|
|
Example:
|
|
|
|
wget https://github.com/k2-fsa/sherpa-onnx/releases/download/speech-enhancement-models/sherpa-onnx-voxlock-int8.tar.bz2
|
|
tar xvf sherpa-onnx-voxlock-int8.tar.bz2
|
|
"""
|
|
|
|
import time
|
|
from pathlib import Path
|
|
|
|
import numpy as np
|
|
import sherpa_onnx
|
|
import soundfile as sf
|
|
|
|
|
|
def create_speech_denoiser():
|
|
enroll_model = "./sherpa-onnx-voxlock-int8/enroll.onnx"
|
|
chunk_model = "./sherpa-onnx-voxlock-int8/chunk_int8.onnx"
|
|
if not Path(enroll_model).is_file() or not Path(chunk_model).is_file():
|
|
raise ValueError(
|
|
"Please first download the models from "
|
|
"https://github.com/k2-fsa/sherpa-onnx/releases/tag/speech-enhancement-models"
|
|
)
|
|
|
|
config = sherpa_onnx.OnlineSpeechDenoiserConfig(
|
|
model=sherpa_onnx.OfflineSpeechDenoiserModelConfig(
|
|
voxlock=sherpa_onnx.OfflineSpeechDenoiserVoxLockModelConfig(
|
|
enroll_model=enroll_model,
|
|
chunk_model=chunk_model,
|
|
voice_sample="",
|
|
),
|
|
debug=False,
|
|
num_threads=1,
|
|
provider="cpu",
|
|
)
|
|
)
|
|
|
|
if not config.validate():
|
|
print(config)
|
|
raise ValueError("Errors in config. Please check previous error logs")
|
|
|
|
return sherpa_onnx.OnlineSpeechDenoiser(config)
|
|
|
|
|
|
def load_audio(filename: str):
|
|
data, sample_rate = sf.read(filename, always_2d=True, dtype="float32")
|
|
samples = np.ascontiguousarray(data[:, 0])
|
|
return samples, sample_rate
|
|
|
|
|
|
def main():
|
|
sd = create_speech_denoiser()
|
|
test_wave = "./sherpa-onnx-voxlock-int8/test_wavs/en-combined.wav"
|
|
if not Path(test_wave).is_file():
|
|
raise ValueError(
|
|
f"{test_wave} does not exist. You can download it from "
|
|
"https://github.com/k2-fsa/sherpa-onnx/releases/tag/speech-enhancement-models"
|
|
)
|
|
|
|
samples, sample_rate = load_audio(test_wave)
|
|
frame_shift = sd.frame_shift_in_samples
|
|
output = []
|
|
|
|
start = time.time()
|
|
for offset in range(0, len(samples), frame_shift):
|
|
chunk = samples[offset : offset + frame_shift]
|
|
denoised = sd(chunk, sample_rate)
|
|
output.append(np.asarray(denoised.samples, dtype=np.float32))
|
|
|
|
output.append(np.asarray(sd.flush().samples, dtype=np.float32))
|
|
end = time.time()
|
|
|
|
elapsed_seconds = end - start
|
|
audio_duration = len(samples) / sample_rate
|
|
real_time_factor = elapsed_seconds / audio_duration
|
|
|
|
enhanced = np.concatenate(output) if output else np.empty(0, dtype=np.float32)
|
|
|
|
sf.write("./voxlock-enhanced-online-16k.wav", enhanced, sd.sample_rate)
|
|
print("Saved to ./voxlock-enhanced-online-16k.wav")
|
|
print(f"Elapsed seconds: {elapsed_seconds:.3f}")
|
|
print(f"Audio duration in seconds: {audio_duration:.3f}")
|
|
print(f"RTF: {elapsed_seconds:.3f}/{audio_duration:.3f} = {real_time_factor:.3f}")
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|