initial commit

README copy.md

@@ -0,0 +1,13 @@
+---
+title: Speech To Speech Translation
+emoji: 🏆
+colorFrom: pink
+colorTo: indigo
+sdk: gradio
+sdk_version: 3.36.1
+app_file: app.py
+pinned: false
+duplicated_from: course-demos/speech-to-speech-translation
+---
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -0,0 +1,82 @@
+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")
+
+
+machine_translate_pipe = pipeline("translation", model="Helsinki-NLP/opus-mt-en-fr")
+
+model = SpeechT5ForTextToSpeech.from_pretrained("Sandiago21/speecht5_finetuned_facebook_voxpopuli_french").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)
+
+
+def translate(audio):
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
+    return outputs["text"]
+
+def machine_translate(text):
+    outputs = machine_translate_pipe(text)
+    return outputs[0]['translation_text']
+
+def synthesise(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)
+    translated_text = machine_translate(translated_text)
+    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 French. Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and Microsoft's
+[SpeechT5 TTS finetuned for french by Sandiago](https://huggingface.co/Sandiago21/speecht5_finetuned_facebook_voxpopuli_french) model for text-to-speech:
+
+![Cascaded STST](https://huggingface.co/datasets/huggingface-course/audio-course-images/resolve/main/s2st_cascaded.png "Diagram of cascaded speech to speech translation")
+"""
+
+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"),
+    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"),
+    examples=[["./example.wav"]],
+    title=title,
+    description=description,
+)
+
+with demo:
+    gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "Audio File"])
+
+demo.launch()

packages.txt

@@ -0,0 +1 @@
+ffmpeg

requirements.txt

@@ -0,0 +1,4 @@
+torch
+git+https://github.com/huggingface/transformers
+datasets
+sentencepiece