truncate audio to 10s and warn (#123)

Truncate to 10s for model. Warn if going to truncate.

---------

Co-authored-by: Milad Alizadeh <git@mil.ad>

Dockerfile

@@ -16,10 +16,8 @@ RUN apt-get update && apt-get install -y \
     && rm -rf /var/lib/apt/lists/* \
     && git lfs install
 
-# TODO: Pin esp-research and esp-data revisions
-# TODO: remove hf-app branch once merged
 RUN --mount=type=secret,id=GH_TOKEN,mode=0444,required=true \
-    git clone -b update-hf-app --single-branch --depth 1 https://$(cat /run/secrets/GH_TOKEN)@github.com/earthspecies/esp-research.git /app/esp-research && \
+    git clone --single-branch --depth 1 https://$(cat /run/secrets/GH_TOKEN)@github.com/earthspecies/esp-research.git /app/esp-research && \
     git clone --single-branch --depth 1 https://$(cat /run/secrets/GH_TOKEN)@github.com/earthspecies/esp-data.git /app/esp-data
 
 # esp-research installs esp-data from gcloud artifact registry, which is not

app.py

@@ -22,8 +22,8 @@ ASSETS_DIR = APP_DIR / "assets"
 # TODO: Set these values carefully later.
 SAMPLE_RATE = 16000
 MIN_AUDIO_DURATION: float = 0.5  # seconds
-MAX_AUDIO_DURATION: float = 60.0  # seconds
 MAX_HISTORY_TURNS = 3
+MODEL_MAX_AUDIO_DURATION: float = 10.0  # seconds – model was trained on 10 s clips
 
 assert torch.cuda.is_available(), "CUDA is required to run this app"
 DEVICE = "cuda"
@@ -39,7 +39,7 @@ logger.info("Model loaded successfully")
 
 
 def validate_audio(audio_path: str) -> None:
-    """Validate that the audio file meets duration requirements.
+    """Validate that the audio file meets the minimum duration requirement.
 
     Parameters
     ----------
@@ -49,15 +49,35 @@ def validate_audio(audio_path: str) -> None:
     Raises
     ------
     Error
-        If the audio duration is outside [`MIN_AUDIO_DURATION`,
-        `MAX_AUDIO_DURATION`].
+        If the audio duration is shorter than `MIN_AUDIO_DURATION`.
     """
     info = sf.info(audio_path)
     duration = info.duration
     if duration < MIN_AUDIO_DURATION:
         raise gr.Error(f"Audio duration must be at least {MIN_AUDIO_DURATION} seconds.")
-    if duration > MAX_AUDIO_DURATION:
-        raise gr.Error(f"Audio duration must be at most {MAX_AUDIO_DURATION} seconds.")
+
+
+def check_truncation_warning(audio_path: str | None) -> dict:
+    """Return a visibility update for the truncation warning banner.
+
+    Parameters
+    ----------
+    audio_path : str | None
+        Path to the uploaded audio file, or ``None`` when audio is cleared.
+
+    Returns
+    -------
+    dict
+        A `gr.update` with ``visible=True`` when the audio exceeds
+        `MODEL_MAX_AUDIO_DURATION`, otherwise ``visible=False``.
+    """
+    if not audio_path:
+        return gr.update(visible=False)
+    try:
+        duration = sf.info(audio_path).duration
+    except Exception:
+        return gr.update(visible=False)
+    return gr.update(visible=duration > MODEL_MAX_AUDIO_DURATION)
 
 
 @spaces.GPU
@@ -93,6 +113,10 @@ def get_response(chatbot_history: list[dict], audio_input: str) -> list[dict]:
             audio_np = librosa.resample(
                 y=audio_np, orig_sr=sr, target_sr=SAMPLE_RATE, res_type="kaiser_best", scale=True
             )
+        max_samples = int(SAMPLE_RATE * MODEL_MAX_AUDIO_DURATION)
+        if len(audio_np) > max_samples:
+            audio_np = audio_np[:max_samples]
+
         audio_tensor = torch.from_numpy(audio_np).to(DEVICE)
 
         # Build chat-format messages for model.generate().
@@ -271,6 +295,15 @@ def main() -> gr.Blocks:
                     )
 
                 with gr.Column(visible=True) as upload_section:
+                    truncation_warning = gr.HTML(
+                        '<div style="background:#FEFCE8; border:1px solid #F5E6A3;'
+                        " border-radius:8px; padding:10px 14px; color:#92820E;"
+                        ' font-size:14px;">'
+                        f"&#9432; Only the first {MODEL_MAX_AUDIO_DURATION:.0f}"
+                        " seconds will be analyzed. Trim to the most relevant"
+                        " section.</div>",
+                        visible=False,
+                    )
                     audio_input = gr.Audio(
                         container=True,
                         interactive=True,
@@ -355,7 +388,7 @@ def main() -> gr.Blocks:
                     return updated_history, ""
 
                 clear_button = gr.ClearButton(
-                    components=[chatbot, chat_input, audio_input, plotter],
+                    components=[chatbot, chat_input, audio_input, plotter, truncation_warning],
                     visible=False,
                 )
 
@@ -393,6 +426,10 @@ def main() -> gr.Blocks:
                         chat,
                         plotter,
                     ],
+                ).then(
+                    fn=check_truncation_warning,
+                    inputs=[audio_input],
+                    outputs=[truncation_warning],
                 ).then(
                     fn=make_spectrogram_figure,
                     inputs=[audio_input],