Spaces:
Running
on
Zero
Running
on
Zero
Testing
Browse files- app.py +5 -6
- requirements.txt +2 -1
app.py
CHANGED
|
@@ -9,7 +9,8 @@ device = "cuda" if torch.cuda.is_available() else "cpu"
|
|
| 9 |
model_name = "Hemg/human-emotion-detection"
|
| 10 |
feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(model_name)
|
| 11 |
model = Wav2Vec2ForSequenceClassification.from_pretrained(model_name)
|
| 12 |
-
|
|
|
|
| 13 |
|
| 14 |
def preprocess_audio(audio):
|
| 15 |
waveform, sampling_rate = torchaudio.load(audio)
|
|
@@ -20,7 +21,7 @@ def preprocess_audio(audio):
|
|
| 20 |
def inference(audio):
|
| 21 |
example = preprocess_audio(audio)
|
| 22 |
inputs = feature_extractor(example['speech'], sampling_rate=16000, return_tensors="pt", padding=True)
|
| 23 |
-
inputs = inputs
|
| 24 |
with torch.no_grad():
|
| 25 |
logits = model(**inputs).logits
|
| 26 |
predicted_ids = torch.argmax(logits, dim=-1)
|
|
@@ -29,11 +30,9 @@ def inference(audio):
|
|
| 29 |
|
| 30 |
iface = gr.Interface(fn=inference,
|
| 31 |
inputs=gr.Audio(type="filepath"),
|
| 32 |
-
outputs=[gr.Label(label="Predicted Sentiment"),
|
| 33 |
-
gr.JSON(label="Logits"),
|
| 34 |
-
gr.JSON(label="Predicted ID")],
|
| 35 |
title="Audio Sentiment Analysis",
|
| 36 |
description="Upload an audio file or record one to analyze sentiment.")
|
| 37 |
|
| 38 |
|
| 39 |
-
iface.launch(
|
|
|
|
| 9 |
model_name = "Hemg/human-emotion-detection"
|
| 10 |
feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(model_name)
|
| 11 |
model = Wav2Vec2ForSequenceClassification.from_pretrained(model_name)
|
| 12 |
+
print(device)
|
| 13 |
+
|
| 14 |
|
| 15 |
def preprocess_audio(audio):
|
| 16 |
waveform, sampling_rate = torchaudio.load(audio)
|
|
|
|
| 21 |
def inference(audio):
|
| 22 |
example = preprocess_audio(audio)
|
| 23 |
inputs = feature_extractor(example['speech'], sampling_rate=16000, return_tensors="pt", padding=True)
|
| 24 |
+
inputs = inputs # Move inputs to GPU
|
| 25 |
with torch.no_grad():
|
| 26 |
logits = model(**inputs).logits
|
| 27 |
predicted_ids = torch.argmax(logits, dim=-1)
|
|
|
|
| 30 |
|
| 31 |
iface = gr.Interface(fn=inference,
|
| 32 |
inputs=gr.Audio(type="filepath"),
|
| 33 |
+
outputs=[gr.Label(label="Predicted Sentiment")],
|
|
|
|
|
|
|
| 34 |
title="Audio Sentiment Analysis",
|
| 35 |
description="Upload an audio file or record one to analyze sentiment.")
|
| 36 |
|
| 37 |
|
| 38 |
+
iface.launch()
|
requirements.txt
CHANGED
|
@@ -1,4 +1,5 @@
|
|
| 1 |
torch
|
| 2 |
transformers
|
| 3 |
accelerate
|
| 4 |
-
torchaudio
|
|
|
|
|
|
| 1 |
torch
|
| 2 |
transformers
|
| 3 |
accelerate
|
| 4 |
+
torchaudio
|
| 5 |
+
accelerate
|