# Copyright (c) 2024 Alibaba Inc (authors: Shengkui Zhao, Zexu Pan) # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. import torch import soundfile as sf import gradio as gr from clearvoice import ClearVoice import random import os # SDK模型下载 from modelscope import snapshot_download snapshot_download('iic/ClearerVoice-Studio', local_dir='checkpoints/') def fn_clearvoice_se(input_wav, model): if model == "FRCRN_16000Hz": myClearVoice = ClearVoice(task='speech_enhancement', model_names=['FRCRN_SE_16K']) fs = 16000 elif model == "MossFormerGAN_16000Hz": myClearVoice = ClearVoice(task='speech_enhancement', model_names=['MossFormerGAN_SE_16K']) fs = 16000 else: myClearVoice = ClearVoice(task='speech_enhancement', model_names=['MossFormer2_SE_48K']) fs = 48000 output_wav_dict = myClearVoice(input_path=input_wav, online_write=False) if isinstance(output_wav_dict, dict): key = next(iter(output_wav_dict)) output_wav = output_wav_dict[key] else: output_wav = output_wav_dict sf.write('enhanced.wav', output_wav[0,:], fs) return 'enhanced.wav' def fn_clearvoice_ss(input_wav): myClearVoice = ClearVoice(task='speech_separation', model_names=['MossFormer2_SS_16K']) output_wav_dict = myClearVoice(input_path=input_wav, online_write=False) if isinstance(output_wav_dict, dict): key = next(iter(output_wav_dict)) output_wav_list = output_wav_dict[key] output_wav_s1 = output_wav_list[0] output_wav_s2 = output_wav_list[1] else: output_wav_list = output_wav_dict output_wav_s1 = output_wav_list[0] output_wav_s2 = output_wav_list[1] sf.write('separated_s1.wav', output_wav_s1[0,:], 16000) sf.write('separated_s2.wav', output_wav_s2[0,:], 16000) return "separated_s1.wav", "separated_s2.wav" def find_mp4_files(directory): mp4_files = [] # Walk through the directory and its subdirectories for root, dirs, files in os.walk(directory): for file in files: # Check if the file ends with .mp4 if file.endswith(".mp4") and file[:3] == 'est': mp4_files.append(os.path.join(root, file)) return mp4_files def fn_clearvoice_tse(input_video): myClearVoice = ClearVoice(task='target_speaker_extraction', model_names=['AV_MossFormer2_TSE_16K']) #output_wav_dict = print(f'input_video: {input_video}') myClearVoice(input_path=input_video, online_write=True, output_path='path_to_output_videos_tse') output_list = find_mp4_files(f'path_to_output_videos_tse/AV_MossFormer2_TSE_16K/{os.path.basename(input_video).split(".")[0]}/') return output_list def fn_clearvoice_sr(input_wav, apply_se): wavname = input_wav.split('/')[-1] myClearVoice = ClearVoice(task='speech_super_resolution', model_names=['MossFormer2_SR_48K']) fs = 48000 if apply_se: new_wavname = wavname.replace('.wav', str(random.randint(0,100))+'.wav') myClearVoice_se = ClearVoice(task='speech_enhancement', model_names=['MossFormer2_SE_48K']) output_wav_dict = myClearVoice_se(input_path=input_wav, online_write=True, output_path=new_wavname) input_wav = new_wavname output_wav_dict = myClearVoice(input_path=input_wav, online_write=False) if isinstance(output_wav_dict, dict): key = next(iter(output_wav_dict)) output_wav = output_wav_dict[key] else: output_wav = output_wav_dict sf.write('enhanced_high_res.wav', output_wav[0,:], fs) return 'enhanced_high_res.wav' demo = gr.Blocks() se_demo = gr.Interface( fn=fn_clearvoice_se, inputs = [ gr.Audio(label="Input Audio", type="filepath"), gr.Dropdown( ["FRCRN_16000Hz", "MossFormerGAN_16000Hz", "MossFormer2_48000Hz"], value="MossFormer2_48000Hz", multiselect=False, info="Choose a speech enhancement model." ), ], outputs = [ gr.Audio(label="Output Audio", type="filepath"), ], title = "ClearerVoice-Studio: 语音增强（语音降噪）", description = ("ClearerVoice-Studio ([Github仓库](https://github.com/modelscope/ClearerVoice-Studio)) 是一个AI驱动的语音处理工具平台，语音增强模块可从背景噪声中提取清晰语音，提升语音质量。它支持16 kHz和48 kHz的音频输出以及多种音频格式。 上传您的音频、或者点击其中一个示例，来尝试一下吧。" "ClearerVoice-Studio ([Github Repo](https://github.com/modelscope/ClearerVoice-Studio)) is AI-powered and extracts clear speech from background noise for enhanced speech quality. It supports both 16 kHz and 48 kHz audio outputs. " "To try it, simply upload your audio, or click one of the examples."), article = ("

FRCRN: Boosting Feature Representation Using Frequency Recurrence for Monaural Speech Enhancement (ICASSP 2022)

" "

MossFormer2: Combining Transformer and RNN-Free Recurrent Network for Enhanced Time-Domain Monaural Speech Separation (ICASSP 2024)

"), examples = [ ["examples/mandarin_speech_16kHz.wav", "FRCRN_16000Hz"], ["examples/mandarin_speech_16kHz.wav", "MossFormerGAN_16000Hz"], ["examples/english_speech_48kHz.wav", "MossFormer2_48000Hz"], ], cache_examples = False, ) ss_demo = gr.Interface( fn=fn_clearvoice_ss, inputs = [ gr.Audio(label="Input Audio", type="filepath"), ], outputs = [ gr.Audio(label="Output Audio", type="filepath"), gr.Audio(label="Output Audio", type="filepath"), ], title = "ClearerVoice-Studio: 语音分离", description = ("ClearerVoice-Studio ([Github仓库](https://github.com/modelscope/ClearerVoice-Studio)) 是一个AI驱动的语音处理工具平台，语音分离模块可从混合音频中分离出单独的语音。它支持16 kHz的音频输出和两个输出流。上传您的音频、或者点击其中一个示例，来尝试一下吧。" "ClearerVoice-Studio ([Github Repo](https://github.com/modelscope/ClearerVoice-Studio)) is powered by AI and separates individual speech from mixed audio. It supports 16 kHz and two output streams. " "To try it, simply upload your audio, or click one of the examples. "), article = ("

MossFormer: Pushing the Performance Limit of Monaural Speech Separation using Gated Single-Head Transformer with Convolution-Augmented Joint Self-Attentions (ICASSP 2023)

" "

MossFormer2: Combining Transformer and RNN-Free Recurrent Network for Enhanced Time-Domain Monaural Speech Separation (ICASSP 2024)

"), examples = [ ['examples/female_female_speech.wav'], ['examples/female_male_speech.wav'], ], cache_examples = False, ) tse_demo = gr.Interface( fn=fn_clearvoice_tse, inputs = [ gr.Video(label="Input Video"), ], outputs = [ gr.Gallery(label="Output Video List") ], title = "ClearerVoice-Studio: 音视频说话人提取", description = ("ClearerVoice-Studio ([Github仓库](https://github.com/modelscope/ClearerVoice-Studio)) 是一个AI驱动的语音处理工具平台，音视频说话人提取模块通过面部识别从多语者视频中提取每个说话者的语音以及图像。上传您的视频、或者点击其中一个示例，来尝试一下吧。" "ClearerVoice-Studio ([Github Repo](https://github.com/modelscope/ClearerVoice-Studio)) is AI-powered and extracts each speaker's voice from a multi-speaker video using facial recognition. " "To try it, simply upload your video, or click one of the examples. "), examples = [ ['examples/001.mp4'], ['examples/002.mp4'], ], cache_examples = False, ) sr_demo = gr.Interface( fn=fn_clearvoice_sr, inputs = [ gr.Audio(label="Input Audio", type="filepath"), gr.Checkbox(label="Apply Speech Enhancement", value=True), ], outputs = [ gr.Audio(label="Output Audio", type="filepath"), ], title = "ClearerVoice-Studio: 语音超分辨率", description = ("ClearerVoice-Studio ([Github仓库](https://github.com/modelscope/ClearerVoice-Studio)) 是一个AI驱动的语音处理工具平台，语音超分辨率模块可将低分辨率音频（有效采样率≥16 kHz）转换为清晰、高清的48 kHz音频。它支持大多数音频格式。" "ClearerVoice-Studio ([Github Repo](https://github.com/modelscope/ClearerVoice-Studio)) is AI-powered and transform low-resolution audio (effective sampling rate ≥ 16 kHz) into crystal-clear, high-resolution audio at 48 kHz. It supports most of audio types. " "To try it, simply upload your audio, or click one of the examples. "), article = ("

FRCRN: Boosting Feature Representation Using Frequency Recurrence for Monaural Speech Enhancement (ICASSP 2022)

" "

MossFormer2: Combining Transformer and RNN-Free Recurrent Network for Enhanced Time-Domain Monaural Speech Separation, (ICASSP 2024)

" "

HiFi-SR: A Unified Generative Transformer-Convolutional Adversarial Network for High-Fidelity Speech Super-Resolution, (ICASSP 2025)

" ), examples = [ ["examples/mandarin_speech_16kHz.wav", True], ["examples/LJSpeech-001-0001-22k.wav", True], ["examples/LibriTTS_986_129388_24k.wav", True], ["examples/english_speech_48kHz.wav", True], ], cache_examples = False, ) if __name__ == '__main__': with demo: #gr.TabbedInterface([sr_demo], ["Task 4: Speech Super Resolution"]) gr.TabbedInterface([se_demo, ss_demo, sr_demo, tse_demo], ["Task 1: Speech Enhancement", "Task 2: Speech Separation", "Task 3: Speech Super Resolution", "Task 4: Audio-Visual Speaker Extraction"]) demo.launch()