feat: start adding server code

generate_manual.py

@@ -0,0 +1,38 @@
+import torchaudio
+import time
+from audiocraft.models.musicgen import MusicGen
+from audiocraft.data.audio import audio_write
+
+MODEL_NAME = "facebook/musicgen-large"
+MUSIC_DURATION_SECONDS = 60
+
+print("obtaining model...")
+
+model = MusicGen.get_pretrained(MODEL_NAME)
+model.set_generation_params(duration=MUSIC_DURATION_SECONDS)
+descriptions = [
+    "gentle, calming lo-fi beats that helps with studying and focusing",
+    "calm, piano lo-fi beats to help with studying and focusing",
+    "gentle lo-fi hip-hop to relax to",
+    "gentle, quiet synthwave lo-fi beats",
+    "morning lo-fi beats",
+]
+
+print("model obtained. generating audio...")
+
+a = time.time()
+wav = model.generate(descriptions)
+b = time.time()
+
+print(f"audio generated. took {b - a} seconds.")
+
+for idx, one_wav in enumerate(wav):
+    # Will save under {idx}.wav, with loudness normalization at -14 db LUFS.
+    audio_write(
+        f"{idx}",
+        one_wav.cpu(),
+        model.sample_rate,
+        format="mp3",
+        strategy="loudness",
+        loudness_compressor=True,
+    )