Test huggingface

app.py

@@ -1,20 +1,26 @@
 import gradio as gr
 import torch
 import scipy.io.wavfile as wavfile
-from transformers import AutoProcessor, SeamlessM4TModel
+from transformers import AutoProcessor, SeamlessM4TModel, pipeline
 
-tokenizer = AutoProcessor.from_pretrained("facebook/hf-seamless-m4t-medium")
-model = SeamlessM4TModel.from_pretrained("facebook/hf-seamless-m4t-medium")
+# tokenizer = AutoProcessor.from_pretrained("facebook/hf-seamless-m4t-medium")
+# model = SeamlessM4TModel.from_pretrained("facebook/hf-seamless-m4t-medium")
 
-text = "some example text in the English language"
+# text = "some example text in the English language"
 
-def greet(text):
-    inputs = tokenizer(text, return_tensors="pt")
-    with torch.no_grad():
-        output = model(**inputs, decoder_input_ids=inputs["input_ids"]).waveform
-        out = output[0]
-        wavfile.write("tmp.wav", rate=16000, data=out)
-        return open("tmp.wav", "rb").read()
+# def greet(text):
+#     inputs = tokenizer(text, return_tensors="pt")
+#     with torch.no_grad():
+#         output = model(**inputs, decoder_input_ids=inputs["input_ids"]).waveform
+#         out = output[0]
+#         wavfile.write("tmp.wav", rate=16000, data=out)
+#         return open("tmp.wav", "rb").read()
 
-iface = gr.Interface(fn=greet, inputs="text", outputs="audio")
+def stt(audio):
+    print(audio)
+    br, data = audio
+    tscrb = pipeline("automatic-speech-recognition", model="facebook/hubert-large-ls960-ft")
+    return tscrb(data)
+
+iface = gr.Interface(fn=stt, inputs="audio", outputs="text")
 iface.launch()