Spaces:
Sleeping
Sleeping
File size: 3,531 Bytes
c03d869 132bef4 8ed9d92 35fee4b 8ed9d92 c03d869 132bef4 8ed9d92 c03d869 35fee4b c03d869 35fee4b c03d869 35fee4b c03d869 e1470c7 c03d869 35fee4b c03d869 35fee4b c03d869 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 |
import gradio as gr
import numpy as np
import torch
from datasets import load_dataset
from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
device = "cuda:0" if torch.cuda.is_available() else "cpu"
# load speech translation checkpoint
asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
# load text-to-speech checkpoint and speaker embeddings
#processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
#"ihanif/speecht5_finetuned_voxpopuli_lt"
#model_id = "sanchit-gandhi/speecht5_tts_vox_nl"
model_id = "ihanif/speecht5_finetuned_voxpopuli_lt"
processor = SpeechT5Processor.from_pretrained(model_id)
#model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
#vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
model = SpeechT5ForTextToSpeech.from_pretrained(model_id).to(device)
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
replacements = [
("à", "a"),
("ą", "a"),
("ç", "c"),
("č", "c"),
("è", "e"),
("ë", "e"),
("ė", "e"),
("ę", "e"),
("í", "i"),
("ï", "i"),
("į", "i"),
("ö", "o"),
("š", "s"),
("ü", "u"),
("ū", "u"),
("ų", "u"),
("ž", "z"),
]
def cleanup_text(text):
for src, dst in replacements:
text = text.replace(src, dst)
return text
def translate(audio):
outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate", "language": "lt"})
return outputs["text"]
def synthesise(text):
text = cleanup_text(text)
inputs = processor(text=text, return_tensors="pt")
speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
return speech.cpu()
def speech_to_speech_translation(audio):
translated_text = translate(audio)
synthesised_speech = synthesise(translated_text)
synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16)
return 16000, synthesised_speech
title = "Cascaded STST"
description = """
Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in Lithuanian. Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and Microsoft's
[SpeechT5 TTS](https://huggingface.co/microsoft/speecht5_tts) model for text-to-speech:
"""
demo = gr.Blocks()
mic_translate = gr.Interface(
fn=speech_to_speech_translation,
inputs=gr.Audio(source="microphone", type="filepath"),
outputs=[gr.Audio(label="Generated Speech", type="numpy"), gr.outputs.Textbox()],
title=title,
description=description,
)
file_translate = gr.Interface(
fn=speech_to_speech_translation,
inputs=gr.Audio(source="upload", type="filepath"),
outputs=[gr.Audio(label="Generated Speech", type="numpy"), gr.outputs.Textbox()],
examples=[["./example.wav"]],
title=title,
description=description,
)
with demo:
gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "Audio File"])
demo.launch()
|