test

app.py

@@ -1,5 +1,5 @@
 #from transformers import pipeline
-import gradio as gr
+#import gradio as gr
 #import nemo.collections.asr as nemo_asr
 #import gradio
 
@@ -33,5 +33,57 @@ gr.Interface(
 ).launch(share=True)
 '''
 #gr.load("models/nvidia/parakeet-ctc-1.1b").launch()
-gr.load("models/openai/whisper-medium.en").launch()
-#gr.load("models/nvidia/stt_en_conformer_ctc_small").launch(share=True)
+#gr.load("models/openai/whisper-medium.en").launch()
+#gr.load("models/nvidia/stt_en_conformer_ctc_small").launch(share=True)
+
+
+import os
+os.system("""pip install nemo_toolkit['all']""")
+
+import nemo.collections.asr as nemo_asr
+from transformers import pipeline
+import numpy as np
+import gradio as gr
+import librosa
+from scipy.io.wavfile import write
+
+def respond(message, chat_history):
+  bot_message = message
+  chat_history.append((message, bot_message))
+  return "", chat_history
+
+def transcribe(audio):
+  sr, y = audio
+  audio_name = "resampled_audio.wav"
+  resampled_audio = librosa.resample(y=y.astype("float"), orig_sr=sr, target_sr=16000)
+  write(audio_name, 16000, resampled_audio)
+  result = asr_model.transcribe([f"./{audio_name}"])
+  return result[0]
+
+asr_model = nemo_asr.models.EncDecCTCModel.from_pretrained(model_name="nvidia/parakeet-ctc-0.6b")
+
+with gr.Blocks() as demo:
+  with gr.Column():
+    gr.Markdown(
+      """
+      # HKU Canteen VA
+      """)
+    va = gr.Chatbot(container=False)
+
+    with gr.Row(): # text input
+      text_input = gr.Textbox(placeholder="Ask me anything...", container=False, scale=1)
+      submit_btn = gr.Button("Submit", scale=0)
+
+    with gr.Row():  # audio input
+      recording = gr.Microphone(show_download_button=False, container=False)
+
+    with gr.Row(): # button toolbar
+      clear = gr.ClearButton([text_input, va])
+
+  text_input.submit(respond, [text_input, va], [text_input, va], queue=False)
+  submit_btn.click(respond, [text_input, va], [text_input, va], queue=False)
+  # recording.stop_recording(transcribe, [recording], [text_input]).then(respond,s [text_input, va], [text_input, va], queue=False)
+  recording.stop_recording(transcribe, [recording], [text_input])
+
+if __name__ == "__main__":
+    demo.launch(share=True)

packages.txt

@@ -0,0 +1,2 @@
+libsndfile1
+ffmpeg

requirements.txt

@@ -1,10 +1,17 @@
 #transformers
 #torch
-gradio
+#gradio
 #nemo_toolkit
 #hydra-core
 #pytorch_lightning
 #Cython
 #librosa
 #sentencepiece
-#youtokentome
+#youtokentome
+
+torch
+transformers
+numpy
+librosa
+scipy
+Cython