gastonduault
commited on
Commit
•
aada626
1
Parent(s):
a15daf1
update predict example
Browse files- README.md +31 -7
- predict-example.py +15 -30
README.md
CHANGED
@@ -28,18 +28,42 @@ You can find a **GitHub** repository with an interface hosted by a Flask API to
|
|
28 |
## Example Usage
|
29 |
```python
|
30 |
from transformers import Wav2Vec2ForSequenceClassification, Wav2Vec2FeatureExtractor
|
|
|
31 |
import torch
|
32 |
|
33 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
34 |
model = Wav2Vec2ForSequenceClassification.from_pretrained("gastonduault/music-classifier")
|
35 |
feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained("facebook/wav2vec2-large")
|
36 |
|
37 |
-
#
|
38 |
-
audio_path
|
39 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
|
41 |
# Predict
|
42 |
with torch.no_grad():
|
43 |
-
logits = model(
|
44 |
-
predicted_class = torch.argmax(logits
|
45 |
-
|
|
|
|
|
|
|
|
28 |
## Example Usage
|
29 |
```python
|
30 |
from transformers import Wav2Vec2ForSequenceClassification, Wav2Vec2FeatureExtractor
|
31 |
+
import librosa
|
32 |
import torch
|
33 |
|
34 |
+
# Genre mapping corrected to a dictionary
|
35 |
+
genre_mapping = {
|
36 |
+
0: "Electronic",
|
37 |
+
1: "Rock",
|
38 |
+
2: "Punk",
|
39 |
+
3: "Experimental",
|
40 |
+
4: "Hip-Hop",
|
41 |
+
5: "Folk",
|
42 |
+
6: "Chiptune / Glitch",
|
43 |
+
7: "Instrumental",
|
44 |
+
8: "Pop",
|
45 |
+
9: "International",
|
46 |
+
}
|
47 |
+
|
48 |
model = Wav2Vec2ForSequenceClassification.from_pretrained("gastonduault/music-classifier")
|
49 |
feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained("facebook/wav2vec2-large")
|
50 |
|
51 |
+
# Function for preprocessing audio for prediction
|
52 |
+
def preprocess_audio(audio_path):
|
53 |
+
audio_array, sampling_rate = librosa.load(audio_path, sr=16000)
|
54 |
+
return feature_extractor(audio_array, sampling_rate=16000, return_tensors="pt", padding=True)
|
55 |
+
|
56 |
+
# Path to your audio file
|
57 |
+
audio_path = "./Nirvana - Come As You Are.wav"
|
58 |
+
|
59 |
+
# Preprocess audio
|
60 |
+
inputs = preprocess_audio(audio_path)
|
61 |
|
62 |
# Predict
|
63 |
with torch.no_grad():
|
64 |
+
logits = model(**inputs).logits
|
65 |
+
predicted_class = torch.argmax(logits, dim=-1).item()
|
66 |
+
|
67 |
+
# Output the result
|
68 |
+
print(f"song analized:{audio_path}")
|
69 |
+
print(f"Predicted genre: {genre_mapping[predicted_class]}")
|
predict-example.py
CHANGED
@@ -1,47 +1,32 @@
|
|
1 |
from transformers import Wav2Vec2ForSequenceClassification, Wav2Vec2FeatureExtractor
|
2 |
-
from datasets import load_dataset
|
3 |
-
import numpy as np
|
4 |
import librosa
|
5 |
import torch
|
6 |
|
7 |
-
#
|
8 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
|
10 |
-
# Load the dataset
|
11 |
-
dataset = load_dataset("lewtun/music_genres_small")
|
12 |
-
|
13 |
-
# Retrieve the label names
|
14 |
-
genre_mapping = {}
|
15 |
-
for example in dataset["train"]:
|
16 |
-
genre_id = example["genre_id"]
|
17 |
-
genre = example["genre"]
|
18 |
-
if genre_id not in genre_mapping:
|
19 |
-
genre_mapping[genre_id] = genre
|
20 |
-
if len(genre_mapping) == 9:
|
21 |
-
break
|
22 |
-
|
23 |
-
print(f"Loading model from {MODEL_DIR}...\n")
|
24 |
model = Wav2Vec2ForSequenceClassification.from_pretrained("gastonduault/music-classifier")
|
25 |
feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained("facebook/wav2vec2-large")
|
26 |
|
27 |
# Function for preprocessing audio for prediction
|
28 |
-
def preprocess_audio(audio_path
|
29 |
audio_array, sampling_rate = librosa.load(audio_path, sr=16000)
|
30 |
-
|
31 |
-
if len(audio_array) > target_length:
|
32 |
-
audio_array = audio_array[:target_length]
|
33 |
-
else:
|
34 |
-
padding = target_length - len(audio_array)
|
35 |
-
audio_array = np.pad(audio_array, (0, padding), "constant")
|
36 |
-
|
37 |
-
inputs = feature_extractor(audio_array, sampling_rate=16000, return_tensors="pt", padding=True)
|
38 |
-
return inputs
|
39 |
-
|
40 |
|
41 |
# Path to your audio file
|
42 |
audio_path = "./Nirvana - Come As You Are.wav"
|
43 |
|
44 |
-
|
45 |
# Preprocess audio
|
46 |
inputs = preprocess_audio(audio_path)
|
47 |
|
|
|
1 |
from transformers import Wav2Vec2ForSequenceClassification, Wav2Vec2FeatureExtractor
|
|
|
|
|
2 |
import librosa
|
3 |
import torch
|
4 |
|
5 |
+
# Genre mapping corrected to a dictionary
|
6 |
+
genre_mapping = {
|
7 |
+
0: "Electronic",
|
8 |
+
1: "Rock",
|
9 |
+
2: "Punk",
|
10 |
+
3: "Experimental",
|
11 |
+
4: "Hip-Hop",
|
12 |
+
5: "Folk",
|
13 |
+
6: "Chiptune / Glitch",
|
14 |
+
7: "Instrumental",
|
15 |
+
8: "Pop",
|
16 |
+
9: "International",
|
17 |
+
}
|
18 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
model = Wav2Vec2ForSequenceClassification.from_pretrained("gastonduault/music-classifier")
|
20 |
feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained("facebook/wav2vec2-large")
|
21 |
|
22 |
# Function for preprocessing audio for prediction
|
23 |
+
def preprocess_audio(audio_path):
|
24 |
audio_array, sampling_rate = librosa.load(audio_path, sr=16000)
|
25 |
+
return feature_extractor(audio_array, sampling_rate=16000, return_tensors="pt", padding=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
26 |
|
27 |
# Path to your audio file
|
28 |
audio_path = "./Nirvana - Come As You Are.wav"
|
29 |
|
|
|
30 |
# Preprocess audio
|
31 |
inputs = preprocess_audio(audio_path)
|
32 |
|