Upload folder using huggingface_hub

README.md

@@ -1,12 +1,6 @@
 ---
-title: Speech To Speech Translation Italian
-emoji: 🌖
-colorFrom: indigo
-colorTo: yellow
-sdk: gradio
-sdk_version: 3.36.1
+title: speech-to-speech-translation-italian
 app_file: app.py
-pinned: false
+sdk: gradio
+sdk_version: 3.36.0
 ---
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -0,0 +1,98 @@
+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-large-v2", device=device)
+
+# load text-to-speech checkpoint and speaker embeddings
+model_id = "Sandiago21/speecht5_finetuned_voxpopuli_it"  # update with your model id
+# pipe = pipeline("automatic-speech-recognition", model=model_id)
+model = SpeechT5ForTextToSpeech.from_pretrained(model_id)
+vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")
+embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
+speaker_embeddings = torch.tensor(embeddings_dataset[7440]["xvector"]).unsqueeze(0)
+
+checkpoint = "microsoft/speecht5_tts"
+processor = SpeechT5Processor.from_pretrained(checkpoint)
+
+replacements = [
+     ("á", "a"),
+     ("ç", "c"),
+     ("è", "e"),
+     ("ì", "i"),
+     ("í", "i"),
+     ("ò", "o"),
+     ("ó", "o"),
+     ("ù", "u"),
+     ("ú", "u"),
+     ("š", "s"),
+     ("ï", "i"),
+]
+
+def cleanup_text(text):
+    for src, dst in replacements:
+        text = text.replace(src, dst)
+    return text
+
+def synthesize_speech(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 gr.Audio.update(value=(16000, speech.cpu().numpy()))
+
+def translate(audio):
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
+    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 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()

packages.txt

@@ -0,0 +1,2 @@
+ffmpeg
+

requirements.txt

@@ -0,0 +1,6 @@
+torch
+git+https://github.com/huggingface/transformers
+datasets
+torchaudio
+sentencepiece
+