first commit

README.md

@@ -1,10 +1,10 @@
 ---
 title: Speech To Speech Translation
-emoji: 💻
-colorFrom: gray
-colorTo: gray
+emoji: 🏆
+colorFrom: pink
+colorTo: indigo
 sdk: gradio
-sdk_version: 3.41.2
+sdk_version: 3.36.1
 app_file: app.py
 pinned: false
 ---

app.py

@@ -0,0 +1,78 @@
+import gradio as gr
+import numpy as np
+import torch
+import os
+from datasets import load_dataset
+
+from transformers import AutoProcessor, SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
+os.environ["CUDA_VISIBLE_DEVICES"]="7"
+
+# device = "cuda" if torch.cuda.is_available() else "cpu"
+device = "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")
+# model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
+# vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
+
+processor = SpeechT5Processor.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl")
+model = SpeechT5ForTextToSpeech.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl")
+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 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()
+    return speech.to(device)
+
+
+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 English. 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:
+
+![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(share=True)

packages.txt

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

requirements.txt

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