Add audio logging

app.py

@@ -21,6 +21,7 @@ import csv
 import torch
 import gradio as gr
 import numpy as np
+import shutil
 from audiocraft.data.audio_utils import convert_audio
 from audiocraft.data.audio import audio_write, audio_read
 from audiocraft.models import MusicGen
@@ -34,13 +35,14 @@ import huggingface_hub
 from huggingface_hub import Repository
 from datetime import datetime
 
-LOCAL = True
+LOCAL = False
 USE_MIDI = True
 
 # LOGS
 DATASET_REPO_URL = "https://huggingface.co/datasets/soundsauce/soundsauce-logs"
 DATA_FILENAME = "ratings.csv"
 DATA_FILE = os.path.join("data", DATA_FILENAME)
+AUDIO_DIR = os.path.join("data", "audio")
 HF_TOKEN = os.environ.get("HF_TOKEN")
 print("is none?", HF_TOKEN is None)
 
@@ -69,16 +71,22 @@ crops = [(0, 5), (0, 10), (0, 15)]
 selected_melody = ""
 selected_crop = None
 selected_text = ""
+output_file = ""
 
 
 def store_message(message: dict):
-    repo.git_pull()
-    if message:
+    if message and output_file:
+        if not os.path.exists(AUDIO_DIR):
+            os.makedirs(AUDIO_DIR)
+        repo.git_pull()
         with open(DATA_FILE, "a") as csvfile:
             writer = csv.DictWriter(csvfile, fieldnames=message.keys())
             writer.writerow(message)
+
+        filepath = os.path.join(AUDIO_DIR, message["TIME"]) + ".mp3"
+        shutil.copy(output_file, filepath)
         commit_url = repo.push_to_hub()
-        print(commit_url)
+        print("Commited to", commit_url)
 
 
 def _call_nostderr(*args, **kwargs):
@@ -149,6 +157,7 @@ def connect_to_endpoint():
 
 
 def _do_predictions(texts, melodies, duration, progress=False, **gen_kwargs):
+    global output_file
     MODEL.set_generation_params(duration=duration, cfg_coef=5, **gen_kwargs)
     print("new batch", len(texts), texts, [None if m is None else (m[0], m[1].shape) for m in melodies])
     be = time.time()
@@ -166,16 +175,12 @@ def _do_predictions(texts, melodies, duration, progress=False, **gen_kwargs):
             melody = convert_audio(melody, sr, target_sr, target_ac)
             processed_melodies.append(melody)
 
-    if any(m is not None for m in processed_melodies):
-        outputs = MODEL.generate_with_chroma(
-            descriptions=texts,
-            melody_wavs=processed_melodies,
-            melody_sample_rate=target_sr,
-            progress=progress,
-        )
-    else:
-        outputs = MODEL.generate(texts, progress=progress)
-
+    outputs = MODEL.generate_with_chroma(
+        descriptions=texts,
+        melody_wavs=processed_melodies,
+        melody_sample_rate=target_sr,
+        progress=progress,
+    )
     outputs = outputs.detach().float()
 
     out_files = []
@@ -204,9 +209,10 @@ def _do_predictions(texts, melodies, duration, progress=False, **gen_kwargs):
 
         audio_write(
             d_filename, demucs_output, MODEL.sample_rate, strategy="loudness",
-            loudness_headroom_db=16, loudness_compressor=True, add_suffix=False)
+            loudness_headroom_db=16, loudness_compressor=True, add_suffix=False, format="mp3")
         out_files.append(d_filename)
         file_cleaner.add(d_filename)
+        output_file = d_filename
     res = [out_file for out_file in out_files]
     for file in res:
         file_cleaner.add(file)
@@ -266,7 +272,7 @@ def crop_melody(melody_file, fname):
     audio_write(fname, melody, sr, strategy="loudness", loudness_headroom_db=16, loudness_compressor=True, add_suffix=False)
 
 def run_remote_model(text, melody, num_retries=3):
-    global selected_text
+    global selected_text, output_file
     print("Running Audiocraft API model with text", text, "and melody", melody.split("/")[-1])
     result = client.predict(
                     text,	# str  in 'Describe your music' Textbox component
@@ -287,7 +293,7 @@ def run_remote_model(text, melody, num_retries=3):
     # Load wav file, if there is an issue with audiocraft, file will not exist
     try:
         output, sr = audio_read(d_filename)
-    except FileNotFoundError:
+    except RuntimeError:
         print("Audiocraft API failed, trying again...")
         if num_retries == 0:
             print("Audiocraft API failed, returning empty file...")
@@ -308,25 +314,28 @@ def run_remote_model(text, melody, num_retries=3):
     output = output.cpu()
     demucs_output = demucs_output.cpu()
 
-
+    file_cleaner.add(d_filename)
+    d_filename = d_filename.replace(".wav", ".mp3")
     audio_write(
         d_filename, demucs_output, 32000, strategy="loudness",
-        loudness_headroom_db=16, loudness_compressor=True, add_suffix=False)
+        loudness_headroom_db=16, loudness_compressor=True, add_suffix=False, format="mp3")
     file_cleaner.add(d_filename)
     selected_text = text
 
     print("Finished", text)
     print("Tempfiles currently stored: ", len(file_cleaner.files))
+    output_file = d_filename
     return d_filename#, gr.File.update(value=d_filename, visible=True)
 
-def rating_callback(rating):
+def rating_callback(rating: int):
+    timestamp = str(datetime.now())
     rating_data = {
         "TEXT": selected_text,
         "MELODY": selected_melody,
         "CROP": selected_crop,
         "RATING": rating,
         "VERSION": "local" if LOCAL else "api",
-        "TIME": str(datetime.now())
+        "TIME": timestamp
     }
     print(rating_data)
     store_message(rating_data)
@@ -361,6 +370,18 @@ def ui_full(launch_kwargs):
                     with gr.Row():
                         slider = gr.Slider(label="Rating", minimum=0, maximum=10, step=1, value=0, scale=2)
                         submit_button = gr.Button("Submit Rating", scale=1)
+                    with gr.Accordion("Show Example Ratings", open=False):
+                        gr.Markdown("""
+                                    ## Example Ratings
+                                    """)
+                        gr.Audio(label="Rating = 0", value="examples/0-rating.mp3")
+                        gr.Audio(label="Rating = 1", value="examples/1-rating.mp3")
+                        gr.Audio(label="Rating = 2", value="examples/2-rating.mp3")
+                        gr.Audio(label="Rating = 3", value="examples/3-rating.mp3")
+                        gr.Audio(label="Rating = 4", value="examples/4-rating.mp3")
+                        gr.Audio(label="Rating = 5", value="examples/5-rating.mp3")
+
+
 
                     # file_download_no_drum = gr.File(label="Download", visible=False)
                     # gr.Markdown(
@@ -393,6 +414,7 @@ def ui_full(launch_kwargs):
                 ["Enchanting Kalimba Melodies atop Mystical Atmosphere"],
             ],
             inputs=[text],
+            label="Example Inputs",
             outputs=[output_without_drum]#, file_download_no_drum]
         )
 
@@ -408,9 +430,9 @@ if __name__ == "__main__":
         help='IP to listen on for connections to Gradio',
     )
     parser.add_argument("--local", action="store_true", help="Run locally instead of using API")
-    parser.add_argument("--midi", action="store_true", help="Render midi instead of wav")
 
     args = parser.parse_args()
+    LOCAL = args.local
 
     launch_kwargs = {}
     launch_kwargs['server_name'] = args.listen