Update app.py

app.py

@@ -1,72 +1,36 @@
 import gradio as gr
-import numpy as np
-import torch
-from datasets import load_dataset
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor, AutoTokenizer, AutoModelForSeq2SeqLM, AutoModelForTextToWaveform
 
-from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
+# Load your pretrained models
+asr_model = Wav2Vec2ForCTC.from_pretrained("Baghdad99/saad-speech-recognition-hausa-audio-to-text")
+asr_processor = Wav2Vec2Processor.from_pretrained("Baghdad99/saad-speech-recognition-hausa-audio-to-text")
 
+# Load the Hausa translation model
+translation_tokenizer = AutoTokenizer.from_pretrained("Baghdad99/saad-hausa-text-to-english-text")
+translation_model = AutoModelForSeq2SeqLM.from_pretrained("Baghdad99/saad-hausa-text-to-english-text")
 
-device = "cuda:0" if torch.cuda.is_available() else "cpu"
+# Load the Text-to-Speech model
+tts_tokenizer = AutoTokenizer.from_pretrained("Baghdad99/english_voice_tts")
+tts_model = AutoModelForTextToWaveform.from_pretrained("Baghdad99/english_voice_tts")
 
-# load speech translation checkpoint
-asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
+def translate_speech(speech):
+    # Transcribe the speech to text
+    inputs = asr_processor(speech, return_tensors="pt", padding=True)
+    logits = asr_model(inputs.input_values).logits
+    predicted_ids = torch.argmax(logits, dim=-1)
+    transcription = asr_processor.decode(predicted_ids[0])
 
-# load text-to-speech checkpoint and speaker embeddings
-processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
+    # Translate the text
+    translated = translation_model.generate(**translation_tokenizer(transcription, return_tensors="pt", padding=True))
+    translated_text = [translation_tokenizer.decode(t, skip_special_tokens=True) for t in translated]
 
-model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
-vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
+    # Convert the translated text to speech
+    inputs = tts_tokenizer(translated_text, return_tensors='pt')
+    audio = tts_model.generate(inputs['input_ids'])
 
-embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
-speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
+    return audio
 
+# Define the Gradio interface
+iface = gr.Interface(fn=translate_speech, inputs=gr.inputs.Audio(source="microphone"), outputs="audio")
+iface.launch()
 
-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()
-
-
-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()