Add audio spectrogram transformer

app.py

@@ -1,37 +1,40 @@
 import gradio as gr
 from transformers import pipeline
 
-model_name = "juliensimon/wav2vec2-conformer-rel-pos-large-finetuned-speech-commands"
+model_names = [
+    "juliensimon/wav2vec2-conformer-rel-pos-large-finetuned-speech-commands",
+    "MIT/ast-finetuned-speech-commands-v2",
+]
 
-p = pipeline("audio-classification", model=model_name)
 
-
-def process(file):
+def process(file, model_name):
+    p = pipeline("audio-classification", model=model_name)
     pred = p(file)
     return {x["label"]: x["score"] for x in pred}
 
 
 # Gradio inputs
-mic = gr.inputs.Audio(source="microphone", type="filepath", label="Speech input")
+mic = gr.Audio(source="microphone", type="filepath", label="Speech input")
+model_selection = gr.Dropdown(model_names, label="Model selection")
 
 # Gradio outputs
-labels = gr.outputs.Label(num_top_classes=3)
+labels = gr.Label(num_top_classes=3)
 
-description = "This Space showcases a wav2vec2-conformer-rel-pos-large model fine-tuned for audio classification on the speech_commands dataset. \n \n It can spot one of the following keywords: 'Yes', 'No', 'Up', 'Down', 'Left', 'Right', 'On', 'Off', 'Stop', 'Go', 'Zero', 'One', 'Two', 'Three', 'Four', 'Five', 'Six', 'Seven', 'Eight', 'Nine', 'Bed', 'Bird', 'Cat', 'Dog', 'Happy', 'House', 'Marvin', 'Sheila', 'Tree', 'Wow', 'Backward', 'Forward', 'Follow', 'Learn', 'Visual'."
+description = "This Space showcases two audio classification models fine-tuned on the speech_commands dataset:\n\n - wav2vec2-conformer: 97.2% accuracy, added in transformers 4.20.0.\n - audio-spectrogram-transformer: 98.12% accuracy, added in transformers 4.25.1.\n \n They can spot one of the following keywords: 'Yes', 'No', 'Up', 'Down', 'Left', 'Right', 'On', 'Off', 'Stop', 'Go', 'Zero', 'One', 'Two', 'Three', 'Four', 'Five', 'Six', 'Seven', 'Eight', 'Nine', 'Bed', 'Bird', 'Cat', 'Dog', 'Happy', 'House', 'Marvin', 'Sheila', 'Tree', 'Wow', 'Backward', 'Forward', 'Follow', 'Learn', 'Visual'."
 
 iface = gr.Interface(
     theme="huggingface",
     description=description,
     fn=process,
-    inputs=[mic],
+    inputs=[mic, model_selection],
     outputs=[labels],
     examples=[
-        ["backward16k.wav"],
-        ["happy16k.wav"],
-        ["marvin16k.wav"],
-        ["seven16k.wav"],
-        ["stop16k.wav"],
-        ["up16k.wav"],
+        ["backward16k.wav", "MIT/ast-finetuned-speech-commands-v2"],
+        ["happy16k.wav", "MIT/ast-finetuned-speech-commands-v2"],
+        ["marvin16k.wav", "MIT/ast-finetuned-speech-commands-v2"],
+        ["seven16k.wav", "MIT/ast-finetuned-speech-commands-v2"],
+        ["stop16k.wav", "MIT/ast-finetuned-speech-commands-v2"],
+        ["up16k.wav", "MIT/ast-finetuned-speech-commands-v2"],
     ],
     allow_flagging="never",
 )

requirements.txt

@@ -1,3 +1,4 @@
 torch
-transformers
+torchaudio
+transformers>=4.25.1
 librosa