Spaces:
Sleeping
Sleeping
Commit
·
bdf45ad
1
Parent(s):
626a7e2
Refactor generate_midi function and remove commented code
Browse files
app.py
CHANGED
|
@@ -68,52 +68,52 @@ def save_wav(filepath):
|
|
| 68 |
# modified_midi.dump_midi(Path(output_midi_path))
|
| 69 |
|
| 70 |
|
| 71 |
-
|
| 72 |
-
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
|
| 76 |
-
#
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
#
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
|
| 87 |
-
|
| 88 |
-
|
| 89 |
-
|
| 90 |
-
|
| 91 |
-
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
|
| 97 |
-
|
| 98 |
-
#
|
| 99 |
-
|
| 100 |
-
|
| 101 |
-
|
| 102 |
-
|
| 103 |
-
#
|
| 104 |
-
|
| 105 |
-
|
| 106 |
-
#
|
| 107 |
-
|
| 108 |
-
|
| 109 |
-
|
| 110 |
-
|
| 111 |
-
#
|
| 112 |
-
|
| 113 |
-
|
| 114 |
-
|
| 115 |
|
| 116 |
-
|
| 117 |
|
| 118 |
@spaces.GPU(duration=120)
|
| 119 |
def gradio_generate(prompt, temperature, max_length):
|
|
|
|
| 68 |
# modified_midi.dump_midi(Path(output_midi_path))
|
| 69 |
|
| 70 |
|
| 71 |
+
def generate_midi(caption, temperature=0.9, max_len=500):
|
| 72 |
+
device = 'cuda' if torch.cuda.is_available() else 'cpu'
|
| 73 |
+
artifact_folder = 'artifacts'
|
| 74 |
+
|
| 75 |
+
tokenizer_filepath = os.path.join(artifact_folder, "vocab_remi.pkl")
|
| 76 |
+
# Load the tokenizer dictionary
|
| 77 |
+
with open(tokenizer_filepath, "rb") as f:
|
| 78 |
+
r_tokenizer = pickle.load(f)
|
| 79 |
+
|
| 80 |
+
# Get the vocab size
|
| 81 |
+
vocab_size = len(r_tokenizer)
|
| 82 |
+
print("Vocab size: ", vocab_size)
|
| 83 |
+
model = Transformer(vocab_size, 768, 8, 2048, 18, 1024, False, 8, device=device)
|
| 84 |
+
model_path = os.path.join("amaai-lab/text2midi", "pytorch_model.bin")
|
| 85 |
+
model.load_state_dict(torch.load(model_path, map_location=device))
|
| 86 |
+
model.eval()
|
| 87 |
+
tokenizer = T5Tokenizer.from_pretrained("google/flan-t5-base")
|
| 88 |
+
|
| 89 |
+
inputs = tokenizer(caption, return_tensors='pt', padding=True, truncation=True)
|
| 90 |
+
input_ids = nn.utils.rnn.pad_sequence(inputs.input_ids, batch_first=True, padding_value=0)
|
| 91 |
+
input_ids = input_ids.to(device)
|
| 92 |
+
attention_mask =nn.utils.rnn.pad_sequence(inputs.attention_mask, batch_first=True, padding_value=0)
|
| 93 |
+
attention_mask = attention_mask.to(device)
|
| 94 |
+
output = model.generate(input_ids, attention_mask, max_len=max_len,temperature = temperature)
|
| 95 |
+
output_list = output[0].tolist()
|
| 96 |
+
generated_midi = r_tokenizer.decode(output_list)
|
| 97 |
+
generated_midi.dump_midi("output.mid")
|
| 98 |
+
# post_processing("output.mid", "output.mid")
|
| 99 |
+
|
| 100 |
+
|
| 101 |
+
@spaces.GPU(duration=120)
|
| 102 |
+
def gradio_generate(prompt, temperature, max_length):
|
| 103 |
+
# Generate midi
|
| 104 |
+
generate_midi(prompt, temperature, max_length)
|
| 105 |
+
|
| 106 |
+
# Convert midi to wav
|
| 107 |
+
midi_filename = "output.mid"
|
| 108 |
+
save_wav(midi_filename)
|
| 109 |
+
wav_filename = midi_filename.replace(".mid", ".wav")
|
| 110 |
+
|
| 111 |
+
# Read the generated WAV file
|
| 112 |
+
output_wave, samplerate = sf.read(wav_filename, dtype='float32')
|
| 113 |
+
temp_wav_filename = "temp.wav"
|
| 114 |
+
wavio.write(temp_wav_filename, output_wave, rate=16000, sampwidth=2)
|
| 115 |
|
| 116 |
+
return temp_wav_filename, midi_filename # Return both WAV and MIDI file paths
|
| 117 |
|
| 118 |
@spaces.GPU(duration=120)
|
| 119 |
def gradio_generate(prompt, temperature, max_length):
|