Spaces:
Paused
Paused
| import torch | |
| import numpy as np | |
| import soundfile as sf | |
| from transformers import pipeline | |
| from transformers import BarkModel | |
| from transformers import AutoProcessor | |
| device = "cuda:0" if torch.cuda.is_available() else "cpu" | |
| pipe = pipeline( | |
| "automatic-speech-recognition", model="openai/whisper-large-v2", device=device | |
| ) | |
| label = pipeline("audio-classification", model="facebook/mms-lid-126", device=device) | |
| processor = AutoProcessor.from_pretrained("suno/bark") | |
| model = BarkModel.from_pretrained("suno/bark") | |
| model = model.to(device) | |
| synthesised_rate = model.generation_config.sample_rate | |
| def translate(audio_file): | |
| audio, sampling_rate = sf.read(audio_file) | |
| outputs = pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe","language":"chinese"}) | |
| language_prediction = label({"array": audio, "sampling_rate": sampling_rate}) | |
| label_outputs = {} | |
| for pred in language_prediction: | |
| label_outputs[pred["label"]] = pred["score"] | |
| return outputs["text"],label_outputs | |
| def synthesise(text_prompt,voice_preset="v2/zh_speaker_1"): | |
| inputs = processor(text_prompt, voice_preset=voice_preset) | |
| speech_output = model.generate(**inputs.to(device),pad_token_id=10000) | |
| return speech_output | |
| def speech_to_speech_translation(audio,voice_preset="v2/zh_speaker_1"): | |
| translated_text, label_outputs= translate(audio) | |
| synthesised_speech = synthesise(translated_text,voice_preset) | |
| synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16) | |
| return (synthesised_rate , synthesised_speech.T),translated_text,label_outputs | |
| title = "Multilanguage to Chinese(mandarin) Cascaded STST" | |
| description = """ | |
| Demo for cascaded speech-to-speech translation (STST), mapping from source speech in Multilanguage to target speech in Chinese(mandarin). Demo uses OpenAI's [Whisper arge-v2](https://huggingface.co/openai/whisper-large-v2) model for speech translation, and a suno/bark[bark-small](https://huggingface.co/suno/bark) model for text-to-speech: | |
|  | |
| """ | |
| examples = [ | |
| ["./cs-CZ.mp3", None], | |
| ["./de-DE.mp3", None], | |
| ["./es-ES.mp3", None], | |
| ["./fr-FR.mp3", None], | |
| ["./it-IT.mp3", None], | |
| ["./ko-KR.mp3", None], | |
| ["./nl-NL.mp3", None], | |
| ["./pl-PL.mp3", None], | |
| ["./pt-PT.mp3", None], | |
| ["./ru-RU.mp3", None], | |
| ] | |
| import gradio as gr | |
| demo = gr.Blocks() | |
| file_transcribe = gr.Interface( | |
| fn=speech_to_speech_translation, | |
| inputs=gr.Audio(source="upload", type="filepath"), | |
| outputs=[ | |
| gr.Audio(label="Generated Speech", type="numpy"), | |
| gr.Text(label="Transcription"), | |
| gr.Label(label="Language prediction"), | |
| ], | |
| title=title, | |
| description=description, | |
| examples=examples, | |
| ) | |
| mic_transcribe = gr.Interface( | |
| fn=speech_to_speech_translation, | |
| inputs=gr.Audio(source="microphone", type="filepath"), | |
| outputs=[ | |
| gr.Audio(label="Generated Speech", type="numpy"), | |
| gr.Text(label="Transcription"), | |
| gr.Label(label="Language prediction"), | |
| ], | |
| title=title, | |
| description=description, | |
| ) | |
| with demo: | |
| gr.TabbedInterface( | |
| [file_transcribe, mic_transcribe], | |
| ["Transcribe Audio File", "Transcribe Microphone"], | |
| ) | |
| demo.launch(share=True) |