dawn_chorus -> voice_focus_examples

app.py

@@ -104,16 +104,18 @@ with gr.Blocks() as demo:
                                 )
 
                         
-        with gr.Tab("Dataset: Dawn Chorus") as dataset_tab:
+        with gr.Tab("Pick Example") as dataset_tab:
             with gr.Group(elem_classes="panel section-panel"):
                 gr.Markdown("### Input", elem_classes="title")
-                gr.Markdown(open("docs/dawn_chorus.md", "r", encoding="utf-8").read(), elem_classes="tab-description")
+                gr.Markdown(open("docs/example_pick.md", "r", encoding="utf-8").read(), elem_classes="tab-description")
 
                 dataset_dropdown = gr.Dropdown(
-                    choices=ALL_FILES, value="en_00352_i_h_33", label="Sample"
+                    choices=ALL_FILES, value=ALL_FILES[0], label="Sample"
                 )
                 audio_file_from_dataset = gr.Audio(
-                    type="filepath", interactive=False, buttons=["download"], autoplay=False
+                    label="Preview",
+                    autoplay=False,
+                    interactive=False,
                 )
 
         with gr.Tab("Upload local file") as upload_tab:
@@ -285,5 +287,5 @@ purge_tmp_directory(max_age_minutes=0, tmp_dir=APP_TMP_DIR)
 demo.queue()
 demo.launch(
     css=(_CSS_DIR / "styling.css").read_text(encoding="utf-8"),
-    allowed_paths=[APP_TMP_DIR, "/tmp", "/"],
+    allowed_paths=[APP_TMP_DIR],
 )

constants.py

@@ -13,11 +13,8 @@ MINUTES_KEEP: Final = 60
 # All app temp files (spectrograms, audio, etc.) go here so we only purge our own files.
 APP_TMP_DIR: Final = "/tmp/voicefocus"
 
-DATASET_NAME: Final = "ai-coustics/dawn_chorus_en"
+DATASET_NAME: Final = "ai-coustics/voice-focus-examples"
 DEFAULT_SPLIT: Final = "eval"
-MIX_DIR: Final = "mix"
-SPEECH_DIR: Final = "speech"
-TRANS_DIR: Final = "transcripts"
 
 STREAM_EVERY: Final = 0.2
 WARMUP_SECONDS: Final = 2  # seconds before "recording ready" light turns on

docs/dawn_chorus.md

@@ -1 +0,0 @@
-Choose a sample from our open-source [Dawn Chorus evaluation dataset](https://huggingface.co/datasets/ai-coustics/dawn_chorus_en). This dataset includes audio with varying levels of background voice activity, ideal for testing the model performance.

docs/example_pick.md

@@ -0,0 +1 @@
+Select a sample from our open-source [Examples](https://huggingface.co/datasets/ai-coustics/voice-focus-examples). These audio files feature different levels of background voice activity, making them ideal for testing model performance.

hf_dataset_utils.py

@@ -6,7 +6,6 @@ from constants import DATASET_NAME, DEFAULT_SPLIT
 # Load once (HF datasets handles caching; HF_TOKEN / login is used automatically if needed)
 ds = load_dataset(DATASET_NAME, split=DEFAULT_SPLIT)
 ds = ds.cast_column("mix", Audio(sampling_rate=16000, decode=True))
-ds = ds.cast_column("speech", Audio(sampling_rate=16000, decode=True))
 
 ALL_FILES = ds["id"]
 

requirements.txt

@@ -8,6 +8,7 @@ loguru~=0.7
 aic-sdk>=2.0.0
 dotenv
 resampy
+websockets
 whisper-normalizer
 soxr
 datasets