Upload 8 files

.gitattributes

@@ -33,3 +33,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+test_audio/clips/D0420-S1-T01_clip.wav filter=lfs diff=lfs merge=lfs -text
+test_audio/clips/D0420-S2-T01_clip.wav filter=lfs diff=lfs merge=lfs -text
+test_audio/clips/D0420-S3-T01_clip.wav filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,15 +1,41 @@
----
-title: Medical Notes Using Medgemma
-emoji: 🏢
-colorFrom: pink
-colorTo: yellow
-sdk: gradio
-sdk_version: 6.20.0
-python_version: '3.12'
-app_file: app.py
-pinned: false
-license: apache-2.0
-short_description: Convert Medical Conversations into Medical Notes.
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# AI Medical Scribe
+
+Pipeline: audio file -> MedASR (transcription) -> MedGemma 4B, `google/medgemma-4b-it`
+(SOAP note) -> Gradio UI.
+
+Both models are gated on the Hugging Face Hub, so whichever environment runs
+this needs an HF token with access to `google/medasr` and `google/medgemma-4b-it`.
+
+## Setup (local)
+
+```bash
+python3.12 -m venv venv
+source venv/bin/activate
+pip install -r requirements.txt
+hf auth login   # token needs access to the gated google/medasr and google/medgemma-4b-it repos
+```
+
+Model weights are pulled straight from the Hub on first run (no local
+`models/` download step needed anymore). To point at a local copy instead,
+set `MEDASR_MODEL_ID` / `MEDGEMMA_MODEL_ID` to a local directory.
+
+## Run (local)
+
+```bash
+python app.py
+```
+
+Opens a local Gradio app: pick a wav/mp3 file, click Run, see the MedASR
+transcript, the generated SOAP note, and per-step timing.
+
+## Deploying to Hugging Face Spaces (ZeroGPU)
+
+1. Create a Space with SDK "Gradio" and hardware "ZeroGPU".
+2. Push this repo's contents to the Space: `app.py`, `llm.py`,
+   `requirements.txt`, `README.md`, `test_audio/clips/*.wav`.
+3. In the Space's Settings -> Variables and secrets, add a secret `HF_TOKEN`
+   set to a token with access to the gated `google/medasr` and
+   `google/medgemma-4b-it` repos. `huggingface_hub` reads this automatically.
+4. `llm.py`'s `run_pipeline` is decorated with `@spaces.GPU`, so ZeroGPU
+   attaches a GPU only for the duration of that call; both the ASR and
+   SOAP-note models load lazily on first invocation.

app.py

@@ -0,0 +1,499 @@
+"""AI medical scribe: audio -> MedASR transcript -> MedGemma SOAP note."""
+
+import os
+
+import gradio as gr
+
+from llm import run_pipeline
+
+_CLIPS_DIR = os.path.join(
+    os.path.dirname(os.path.abspath(__file__)), "test_audio", "clips"
+)
+
+SAMPLES = {
+    "Sample 1 — Psychiatric referral": os.path.join(
+        _CLIPS_DIR, "D0420-S1-T01_clip.wav"
+    ),
+    "Sample 2 — Mood check-in": os.path.join(_CLIPS_DIR, "D0420-S2-T01_clip.wav"),
+    "Sample 3 — Sleep & nightmares": os.path.join(_CLIPS_DIR, "D0420-S3-T01_clip.wav"),
+}
+
+# Hardcoded light palette only — no Gradio theme / dark mode.
+CUSTOM_CSS = """
+:root, :root.dark, :root .dark, .dark, html, body {
+    color-scheme: light !important;
+
+    --radius: 0.9rem;
+    --bg: #f6f7fb;
+    --bg-glow: radial-gradient(ellipse 70% 40% at 15% -10%, rgba(59,130,246,0.07), transparent),
+               radial-gradient(ellipse 60% 40% at 100% 0%, rgba(139,92,246,0.06), transparent);
+    --card: #ffffff;
+    --card-alt: #f2f4f9;
+    --border: #e3e7f0;
+    --fg: #10131c;
+    --muted: #667085;
+    --primary: #3b82f6;
+    --primary-hover: #2563eb;
+    --shadow: 0 1px 2px rgba(16,24,40,0.06), 0 1px 0 rgba(16,24,40,0.02);
+
+    /* Override Gradio tokens so dark class / system preference cannot darken the UI */
+    --body-background-fill: #f6f7fb !important;
+    --background-fill-primary: #f6f7fb !important;
+    --background-fill-secondary: #ffffff !important;
+    --block-background-fill: #ffffff !important;
+    --block-border-color: #e3e7f0 !important;
+    --border-color-primary: #e3e7f0 !important;
+    --body-text-color: #10131c !important;
+    --body-text-color-subdued: #667085 !important;
+    --neutral-50: #f8fafc !important;
+    --neutral-100: #f1f5f9 !important;
+    --neutral-200: #e2e8f0 !important;
+    --neutral-300: #cbd5e1 !important;
+    --neutral-400: #94a3b8 !important;
+    --neutral-500: #64748b !important;
+    --neutral-600: #475569 !important;
+    --neutral-700: #334155 !important;
+    --neutral-800: #1e293b !important;
+    --neutral-900: #0f172a !important;
+    --neutral-950: #020617 !important;
+    --input-background-fill: #f2f4f9 !important;
+    --input-border-color: #e3e7f0 !important;
+}
+
+* { box-sizing: border-box; }
+
+html, body, gradio-app, .gradio-container, .main, .wrap, .contain,
+#root, .app, .fillable, .dark, .dark body, body.dark {
+    background: #f6f7fb !important;
+    background-color: #f6f7fb !important;
+    background-image: var(--bg-glow) !important;
+    background-attachment: fixed !important;
+    color: #10131c !important;
+    font-family: 'Inter', system-ui, sans-serif !important;
+}
+
+footer { display: none !important; }
+.gradio-container {
+    max-width: 100% !important;
+    width: 100% !important;
+    margin: 0 !important;
+    padding: 0 clamp(1.25rem, 5vw, 4rem) !important;
+    min-height: 100vh !important;
+}
+
+/* Strip Gradio block chrome; only .card should paint a background. */
+.block, .form, .gr-group, div[data-testid="block"], .gradio-container .padded {
+    background: transparent !important;
+    border: none !important;
+    box-shadow: none !important;
+}
+
+#top-bar {
+    display: flex;
+    align-items: center;
+    justify-content: space-between;
+    padding: 1.5rem 0 0 0;
+}
+#brand {
+    display: flex;
+    align-items: center;
+    gap: 0.6rem;
+    font-family: 'Outfit', sans-serif;
+    font-weight: 700;
+    font-size: 1.05rem;
+    color: #10131c !important;
+}
+#brand .logo-dot {
+    width: 10px; height: 10px; border-radius: 3px;
+    background: linear-gradient(135deg, #3b82f6, #8b5cf6);
+    box-shadow: 0 0 10px rgba(59,130,246,0.6);
+}
+
+#app-header {
+    text-align: center;
+    padding: 2.75rem 1rem 2.25rem 1rem;
+}
+#app-header .eyebrow {
+    display: inline-block;
+    font-size: 0.72rem;
+    font-weight: 600;
+    letter-spacing: 0.12em;
+    text-transform: uppercase;
+    color: #2563eb !important;
+    background: rgba(59,130,246,0.12);
+    border: 1px solid rgba(59,130,246,0.3);
+    padding: 0.3rem 0.9rem;
+    border-radius: 999px;
+    margin-bottom: 1.1rem;
+}
+#app-header h1 {
+    font-family: 'Outfit', sans-serif;
+    font-weight: 800;
+    font-size: clamp(2rem, 4vw, 2.9rem);
+    letter-spacing: -0.02em;
+    margin: 0 0 0.6rem 0;
+    color: #10131c !important;
+}
+#app-header p {
+    color: #667085 !important;
+    font-size: 1.02rem;
+    max-width: 640px;
+    margin: 0 auto;
+    line-height: 1.6;
+}
+
+.card {
+    background: #ffffff !important;
+    border: 1px solid #e3e7f0 !important;
+    border-radius: var(--radius) !important;
+    box-shadow: var(--shadow);
+}
+
+#control-row { align-items: stretch; gap: 1rem; }
+
+#sample-panel { padding: 1.25rem !important; margin-bottom: 1rem; }
+#sample-radio .wrap {
+    display: flex !important;
+    flex-wrap: wrap;
+    justify-content: center;
+    gap: 0.75rem;
+    background: transparent !important;
+    border: none !important;
+}
+#sample-radio label {
+    position: relative;
+    background: #f9fafc !important;
+    border: 1.5px solid #e3e7f0 !important;
+    border-radius: 0.65rem !important;
+    padding: 0.7rem 1.1rem 0.7rem 2.15rem !important;
+    font-size: 0.85rem !important;
+    font-weight: 600 !important;
+    text-transform: none !important;
+    letter-spacing: normal !important;
+    color: #10131c !important;
+    box-shadow: 0 1px 2px rgba(16,24,40,0.05);
+    cursor: pointer;
+    transition: background 0.15s ease, border-color 0.15s ease, box-shadow 0.15s ease, transform 0.1s ease;
+}
+/* custom radio dot, drawn in place of the hidden native input */
+#sample-radio label::before {
+    content: "";
+    position: absolute;
+    left: 0.85rem;
+    top: 50%;
+    transform: translateY(-50%);
+    width: 15px;
+    height: 15px;
+    border-radius: 50%;
+    border: 1.5px solid #cbd5e1;
+    background: #ffffff;
+    transition: border-color 0.15s ease, background 0.15s ease, box-shadow 0.15s ease;
+}
+#sample-radio label:hover {
+    border-color: #93c5fd !important;
+    box-shadow: 0 2px 8px rgba(59,130,246,0.12);
+    transform: translateY(-1px);
+}
+#sample-radio label:hover::before { border-color: #3b82f6; }
+#sample-radio label:has(input:checked) {
+    background: #eaf1ff !important;
+    border-color: #3b82f6 !important;
+    color: #1d4ed8 !important;
+    box-shadow: 0 3px 12px rgba(59,130,246,0.25);
+}
+#sample-radio label:has(input:checked)::before {
+    border-color: #3b82f6;
+    background: #3b82f6;
+    box-shadow: inset 0 0 0 3px #ffffff;
+}
+#sample-radio input[type="radio"] { display: none; }
+
+label span, .gr-form label {
+    color: #667085 !important;
+    font-weight: 500 !important;
+    font-size: 0.78rem !important;
+    letter-spacing: 0.03em;
+    text-transform: uppercase;
+}
+
+#audio-upload {
+    background: #ffffff !important;
+    border: 1px dashed #e3e7f0 !important;
+    border-radius: var(--radius) !important;
+    color: #10131c !important;
+    min-height: 220px;
+}
+#audio-upload:hover { border-color: #3b82f6 !important; }
+#audio-upload .audio-container,
+#audio-upload button.svelte-8prmba {
+    width: 100% !important;
+    background: transparent !important;
+}
+#audio-upload .wrap.svelte-1vmd51o {
+    width: 100% !important;
+    height: 100% !important;
+    background: transparent !important;
+    color: #10131c !important;
+    gap: 0.35rem;
+}
+#audio-upload .wrap.svelte-1vmd51o .icon-wrap {
+    color: #3b82f6 !important;
+    width: 34px !important;
+    height: 34px !important;
+    margin-bottom: 0.25rem;
+}
+#audio-upload .wrap.svelte-1vmd51o .or {
+    color: #94a3b8 !important;
+    font-size: 0.85rem !important;
+}
+
+button.primary, #run-btn button {
+    background: linear-gradient(180deg, #2563eb, #3b82f6) !important;
+    border: none !important;
+    border-radius: 0.75rem !important;
+    color: white !important;
+    font-weight: 700 !important;
+    font-size: 1rem !important;
+    letter-spacing: 0.01em;
+    box-shadow: 0 6px 20px rgba(59,130,246,0.35);
+    transition: transform 0.15s ease, box-shadow 0.15s ease;
+}
+button.primary:hover, #run-btn button:hover {
+    transform: translateY(-1px);
+    box-shadow: 0 8px 26px rgba(59,130,246,0.5);
+}
+#run-btn button { width: 100%; height: 3.1rem !important; }
+#run-row { margin-top: 1rem; }
+
+textarea, input[type="text"] {
+    background: #f2f4f9 !important;
+    border: 1px solid #e3e7f0 !important;
+    border-radius: 0.6rem !important;
+    color: #10131c !important;
+    font-family: 'Inter', monospace !important;
+    font-size: 0.92rem !important;
+    line-height: 1.55 !important;
+}
+
+#transcript-panel, #soap-panel, #timing-panel {
+    padding: 1.25rem !important;
+}
+
+.panel-title {
+    font-family: 'Outfit', sans-serif;
+    font-weight: 600;
+    font-size: 1.02rem;
+    color: #10131c !important;
+    margin-bottom: 0.9rem;
+    display: flex;
+    align-items: center;
+    gap: 0.5rem;
+}
+.panel-title .dot {
+    width: 8px; height: 8px; border-radius: 50%;
+}
+.dot-blue { background: #3b82f6; box-shadow: 0 0 8px #3b82f6; }
+.dot-violet { background: #8b5cf6; box-shadow: 0 0 8px #8b5cf6; }
+.dot-emerald { background: #10b981; box-shadow: 0 0 8px #10b981; }
+
+#soap-panel .prose, #soap-panel p, #soap-panel li, #soap-panel strong {
+    color: #10131c !important;
+}
+#soap-panel { min-height: 340px; }
+
+.timing-badges { display: flex; gap: 0.75rem; flex-wrap: wrap; }
+.timing-badge {
+    flex: 1;
+    min-width: 150px;
+    background: #f2f4f9;
+    border: 1px solid #e3e7f0;
+    border-radius: 0.75rem;
+    padding: 0.85rem 1rem;
+}
+.timing-badge .label {
+    font-size: 0.72rem;
+    text-transform: uppercase;
+    letter-spacing: 0.08em;
+    color: #667085;
+    margin-bottom: 0.3rem;
+}
+.timing-badge .value {
+    font-family: 'Outfit', sans-serif;
+    font-size: 1.4rem;
+    font-weight: 600;
+    color: #10131c;
+}
+.timing-badge .value span { font-size: 0.95rem; color: #667085; font-weight: 500; }
+"""
+
+HEAD = """
+<link rel="stylesheet" href="https://fonts.googleapis.com/css2?family=Outfit:wght@500;600;700;800&display=swap">
+<script>
+(() => {
+  const LIGHT = "#f6f7fb";
+  let painting = false;
+  const paint = () => {
+    if (painting) return;
+    painting = true;
+    try {
+      document.documentElement.classList.remove("dark");
+      document.body?.classList.remove("dark");
+      document.querySelectorAll(".dark").forEach((el) => el.classList.remove("dark"));
+      for (const el of [document.documentElement, document.body, document.querySelector("gradio-app")]) {
+        if (!el) continue;
+        el.style.setProperty("background", LIGHT, "important");
+        el.style.setProperty("background-color", LIGHT, "important");
+        el.style.setProperty("color-scheme", "light", "important");
+      }
+    } finally {
+      painting = false;
+    }
+  };
+  const url = new URL(window.location.href);
+  if (url.searchParams.get("__theme") !== "light") {
+    url.searchParams.set("__theme", "light");
+    window.history.replaceState({}, "", url);
+  }
+  paint();
+  document.addEventListener("DOMContentLoaded", paint);
+  new MutationObserver((mutations) => {
+    const needs = mutations.some((m) => {
+      const el = m.target;
+      return el.classList?.contains("dark") || (el.style && !String(el.style.background || "").includes(LIGHT));
+    });
+    if (needs) paint();
+  }).observe(document.documentElement, {
+    attributes: true,
+    attributeFilter: ["class", "style"],
+    subtree: true,
+  });
+})();
+</script>
+"""
+
+TOPBAR_HTML = """
+<div id="top-bar">
+    <div id="brand"><span class="logo-dot"></span>MedScribe</div>
+</div>
+"""
+
+HEADER_HTML = """
+<div id="app-header">
+    <span class="eyebrow">MedASR + MedGemma 4B</span>
+    <h1>AI Medical Scribe</h1>
+    <p>Turn a raw clinical recording into a structured SOAP note.</p>
+</div>
+"""
+
+
+def timing_html(transcription_s: float, generation_s: float) -> str:
+    total = transcription_s + generation_s
+    return f"""
+    <div class="timing-badges">
+        <div class="timing-badge">
+            <div class="label">Transcription</div>
+            <div class="value">{transcription_s:.2f}<span>s</span></div>
+        </div>
+        <div class="timing-badge">
+            <div class="label">SOAP Generation</div>
+            <div class="value">{generation_s:.2f}<span>s</span></div>
+        </div>
+        <div class="timing-badge">
+            <div class="label">Total</div>
+            <div class="value">{total:.2f}<span>s</span></div>
+        </div>
+    </div>
+    """
+
+
+def pick_sample(label):
+    return SAMPLES.get(label)
+
+
+def run(audio_file):
+    if audio_file is None:
+        return (
+            "No audio file selected.",
+            "*Select an audio file and click Run.*",
+            timing_html(0, 0),
+        )
+
+    result = run_pipeline(audio_file)
+    return (
+        result.transcript,
+        result.soap_note,
+        timing_html(result.transcription_seconds, result.generation_seconds),
+    )
+
+
+with gr.Blocks(title="AI Medical Scribe") as demo:
+    gr.HTML(TOPBAR_HTML)
+    gr.HTML(HEADER_HTML)
+
+    with gr.Column(elem_classes=["card"], elem_id="sample-panel"):
+        gr.HTML(
+            '<div class="panel-title"><span class="dot dot-blue"></span>Choose a sample recording</div>'
+        )
+        sample_picker = gr.Radio(
+            choices=list(SAMPLES.keys()),
+            value=None,
+            show_label=False,
+            elem_id="sample-radio",
+        )
+
+    with gr.Column(elem_id="control-row"):
+        audio_input = gr.Audio(
+            label="Audio recording",
+            sources=["upload"],
+            type="filepath",
+            elem_id="audio-upload",
+        )
+        with gr.Row(elem_id="run-row"):
+            run_button = gr.Button(
+                "Run pipeline", variant="primary", elem_id="run-btn", size="lg"
+            )
+
+    with gr.Row():
+        with gr.Column(elem_classes=["card"], elem_id="transcript-panel"):
+            gr.HTML(
+                '<div class="panel-title"><span class="dot dot-blue"></span>MedASR Transcript</div>'
+            )
+            transcript_output = gr.Textbox(
+                show_label=False,
+                lines=15,
+                container=False,
+                placeholder="Transcript will appear here…",
+            )
+        with gr.Column(elem_classes=["card"], elem_id="soap-panel"):
+            gr.HTML(
+                '<div class="panel-title"><span class="dot dot-violet"></span>SOAP Note — MedGemma 4B</div>'
+            )
+            soap_output = gr.Markdown("*SOAP note will appear here…*")
+
+    with gr.Column(elem_classes=["card"], elem_id="timing-panel"):
+        gr.HTML(
+            '<div class="panel-title"><span class="dot dot-emerald"></span>Step Timing</div>'
+        )
+        timing_output = gr.HTML(timing_html(0, 0))
+
+    # Picking a sample loads it into the player (so it can be listened to
+    # immediately); the player's own .change then kicks off the pipeline,
+    # so transcription/generation run while playback is already available.
+    sample_picker.change(fn=pick_sample, inputs=[sample_picker], outputs=[audio_input])
+    audio_input.change(
+        fn=run,
+        inputs=[audio_input],
+        outputs=[transcript_output, soap_output, timing_output],
+    )
+    run_button.click(
+        fn=run,
+        inputs=[audio_input],
+        outputs=[transcript_output, soap_output, timing_output],
+    )
+
+if __name__ == "__main__":
+    demo.launch(
+        css=CUSTOM_CSS,
+        head=HEAD,
+        footer_links=[],
+    )

llm.py

@@ -0,0 +1,118 @@
+"""Audio -> MedASR transcript -> MedGemma 4B SOAP note pipeline."""
+
+import os
+import time
+from dataclasses import dataclass
+
+import spaces
+import torch
+from transformers import AutoModelForImageTextToText, AutoProcessor, pipeline
+
+ASR_MODEL_ID = os.environ.get("MEDASR_MODEL_ID", "google/medasr")
+LLM_MODEL_ID = os.environ.get("MEDGEMMA_MODEL_ID", "google/medgemma-4b-it")
+
+SYSTEM_PROMPT = (
+    "You are a clinical documentation engine. Convert the transcript into a "
+    "SOAP note (Subjective, Objective, Assessment, Plan). Infer which "
+    "statements come from the doctor versus the patient based on context "
+    "(questions, clinical observations vs. symptom descriptions). Do not "
+    "fabricate any detail not present in the transcript. If speaker "
+    "attribution is unclear, mark it as unclear rather than guessing "
+    "confidently."
+)
+
+_asr_pipe = None
+_llm_model = None
+_llm_processor = None
+
+
+def _device_and_dtype():
+    if torch.cuda.is_available():
+        return "cuda", torch.bfloat16
+    if torch.backends.mps.is_available():
+        return "mps", torch.float16
+    return "cpu", torch.float32
+
+
+def get_asr_pipeline():
+    global _asr_pipe
+    if _asr_pipe is None:
+        device, dtype = _device_and_dtype()
+        _asr_pipe = pipeline(
+            "automatic-speech-recognition",
+            model=ASR_MODEL_ID,
+            device=device,
+            dtype=dtype,
+        )
+    return _asr_pipe
+
+
+def get_llm():
+    global _llm_model, _llm_processor
+    if _llm_model is None:
+        device, dtype = _device_and_dtype()
+        _llm_model = AutoModelForImageTextToText.from_pretrained(
+            LLM_MODEL_ID, dtype=dtype, device_map=device
+        )
+        _llm_processor = AutoProcessor.from_pretrained(LLM_MODEL_ID)
+    return _llm_model, _llm_processor
+
+
+def transcribe(audio_path: str) -> str:
+    """Transcribe an audio file to text using MedASR, chunked for long audio."""
+    pipe = get_asr_pipeline()
+    result = pipe(audio_path, chunk_length_s=20, stride_length_s=2)
+    return result["text"].strip()
+
+
+def generate_soap_note(transcript: str) -> str:
+    model, processor = get_llm()
+    messages = [
+        {"role": "system", "content": [{"type": "text", "text": SYSTEM_PROMPT}]},
+        {"role": "user", "content": [{"type": "text", "text": transcript}]},
+    ]
+    inputs = processor.apply_chat_template(
+        messages,
+        add_generation_prompt=True,
+        tokenize=True,
+        return_dict=True,
+        return_tensors="pt",
+    ).to(model.device)
+
+    input_len = inputs["input_ids"].shape[-1]
+
+    with torch.inference_mode():
+        generated_ids = model.generate(
+            **inputs,
+            max_new_tokens=1024,
+            do_sample=False,
+            repetition_penalty=1.3,
+        )
+
+    new_tokens = generated_ids[0][input_len:]
+    return processor.decode(new_tokens, skip_special_tokens=True).strip()
+
+
+@dataclass
+class PipelineResult:
+    transcript: str
+    soap_note: str
+    transcription_seconds: float
+    generation_seconds: float
+
+
+@spaces.GPU(duration=120)
+def run_pipeline(audio_path: str) -> PipelineResult:
+    t0 = time.perf_counter()
+    transcript = transcribe(audio_path)
+    t1 = time.perf_counter()
+
+    soap_note = generate_soap_note(transcript)
+    t2 = time.perf_counter()
+
+    return PipelineResult(
+        transcript=transcript,
+        soap_note=soap_note,
+        transcription_seconds=t1 - t0,
+        generation_seconds=t2 - t1,
+    )

requirements.txt

@@ -0,0 +1,11 @@
+torch==2.13.0
+torchaudio==2.11.0
+transformers==5.14.1
+accelerate==1.14.0
+soundfile==0.14.0
+librosa==0.11.0
+huggingface_hub==1.24.0
+gradio==6.20.0
+spaces
+sentencepiece
+pillow

test_audio/clips/D0420-S1-T01_clip.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3aca81a6750e0e091fa7daec5a3046513f8fdded0dcf4b3cbef45ef86b841434
+size 2880154

test_audio/clips/D0420-S2-T01_clip.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b990a7cf8d577e73ab47cd5580f95ab053feee176fad7dc54b007f1def38b97
+size 2880154

test_audio/clips/D0420-S3-T01_clip.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:163265a5609dd0aec3c322786b5296275166350e708523b2ac9c3e44ae08a797
+size 2880154