extended intro and limited file size to 5MB

docs/intro.md

@@ -1,3 +1,14 @@
-Welcome! Try **ai‑coustics Quail Voice Focus**. This is our real‑time enhancement model designed for speech-to-text, which **isolates the main speaker** and reduces background voice acitivity. Learn more in our [documentation](https://docs.ai-coustics.com/guides/models#quail).
+# Welcome! Try ai‑coustics Quail Voice Focus 2.0
 
-This model is optimized to keep the phonetic details that speech-to-text systems need, so the output may not always sound "prettier," but delivers the most accurate transcriptions.
+Our flagship real‑time enhancement model is engineered specifically for Speech-to-Text (STT). By isolating the foreground speaker and suppressing background chatter, it preserves the critical phonetic details that AI systems need to transcribe accurately.
+
+> **Note:** This model is optimized for machine performance rather than perceptual quality. The output may sound less processed to a listener, but this ensures higher transcription accuracy.
+
+## 🚀 Now Native in LiveKit & Pipecat
+
+Build production-ready voice agents with ai‑coustics audio intelligence available natively in the most widely used real-time infrastructure layers:
+
+- **[LiveKit](https://github.com/livekit/plugins-ai-coustics-python)** — official ai‑coustics plugin for LiveKit
+- **[Pipecat](https://github.com/pipecat-ai/pipecat/blob/main/src/pipecat/audio/filters/aic_filter.py)** — ai‑coustics audio filter for Pipecat
+
+[Learn more in our documentation](https://docs.ai-coustics.com/guides/models#quail-voice-focus-2-0-l-16-khz)

offline_pipeline.py

@@ -14,7 +14,8 @@ def retrieve_audio_information(
     sample_id: str,
     stt_model: str,
 ) -> tuple[str, str, str, str, str, str]:
-    
+    if original_array is None or enhanced_array is None:
+        raise ValueError("Audio arrays are not available.")
     noisy_spec_path = f"/tmp/{sample_id}_noisy_spectrogram.png"
     enhanced_spec_path = f"/tmp/{sample_id}_enhanced_spectrogram.png"
     spec_image(original_array).save(noisy_spec_path)
@@ -33,9 +34,12 @@ def retrieve_audio_information(
 
 
 def denoise_audio(
-    sample_16k: np.ndarray, 
+    sample_16k: np.ndarray,
     enhancement_level: float = 50.0,
 ) -> tuple[np.ndarray | None , tuple[int, np.ndarray]| None]:
+    if sample_16k is None:
+        gr.Warning("No audio loaded. Please upload a file first.")
+        raise ValueError("No audio to enhance.")
     try:
         sdk = SDKWrapper()
         sdk.init_processor(sample_rate=DEFAULT_SR, enhancement_level=float(enhancement_level) / 100.0)
@@ -53,6 +57,9 @@ def load_local_file(
     if not sample_path or not os.path.exists(sample_path):
         gr.Warning("Please upload a valid audio file.")
         raise ValueError("Missing audio sample. Please upload an audio sample or use the microphone input.")
+    if os.path.getsize(sample_path) > 5 * 1024 * 1024:
+        gr.Warning("File size exceeds 5 MB limit. Please upload a smaller file.")
+        raise ValueError("Uploaded file exceeds the 5 MB size limit.")
     new_sample_stem = os.path.splitext(os.path.basename(sample_path))[0]
     y_16k, _ = librosa.load(sample_path, sr=DEFAULT_SR, dtype="float32", mono=True)
     return y_16k, new_sample_stem