Unit 7 - Final working copy

app.py

@@ -2,27 +2,25 @@ 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")
+processor = SpeechT5Processor.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl")
 
-model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
+model = SpeechT5ForTextToSpeech.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl").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)
-
+embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation", streaming = True)
+speaker_embeddings = torch.tensor(next(iter(embeddings_dataset))["xvector"]).unsqueeze(0)
 
 def translate(audio):
-    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "nl"})
+    print(outputs["text"])
     return outputs["text"]
 
 
@@ -31,7 +29,6 @@ def synthesise(text):
     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)
@@ -70,3 +67,4 @@ with demo:
     gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "Audio File"])
 
 demo.launch()
+