Fix: revert to robust Docker SDK, add agents.md, and finalize app.py logic

Dockerfile

@@ -0,0 +1,43 @@
+FROM python:3.10
+
+ENV PYTHONDONTWRITEBYTECODE=1
+ENV PYTHONUNBUFFERED=1
+
+RUN useradd -m -u 1000 user
+WORKDIR /home/user/app
+
+# Install system dependencies
+RUN apt-get update && apt-get install -y \
+    ffmpeg \
+    libsndfile1 \
+    cmake \
+    g++ \
+    git \
+    build-essential \
+    && rm -rf /var/lib/apt/lists/*
+
+# Install build tools manually to ensure they are available for youtokentome
+RUN pip install --no-cache-dir --upgrade pip && \
+    pip install --no-cache-dir \
+        Cython \
+        packaging \
+        setuptools \
+        wheel
+
+# Install youtokentome without isolation for NeMo compatibility
+RUN pip install --no-cache-dir --no-build-isolation youtokentome
+
+# Copy requirements and install
+COPY --chown=user requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Copy everything else
+COPY --chown=user . .
+
+# Permissions
+RUN chmod 777 /home/user/app
+
+USER user
+EXPOSE 7860
+
+CMD ["python", "app.py"]

README.md

@@ -3,7 +3,7 @@ title: RobotsMali ASR Demo
 emoji: 🎙️
 colorFrom: blue
 colorTo: green
-sdk: gradio
+sdk: docker
 app_port: 7860
 pinned: false
 ---

agents.md

@@ -0,0 +1,20 @@
+# RobotsMali ASR Agent
+
+This agent provides Automatic Speech Recognition (ASR) for Bambara using state-of-the-art models from RobotsMali.
+
+## Description
+
+This space implements several models optimized for Bambara language transcription, including:
+- **Soloni V3 (TDT-CTC)**: A hybrid architecture for fast and accurate transcription.
+- **Soloba V3 (CTC)**: A robust CTC-based model.
+
+The agent can take audio input (via upload or internal path) and return a full text transcription.
+
+## Tools
+
+### transcription
+Transcribes an audio file into text.
+- **audio**: The audio file to transcribe (Path or URL).
+- **model_name**: The name of the model to use (default: "Soloni V3 (TDT-CTC)").
+
+Returns: String containing the transcription.