voice-vibecoder 2.22.0__tar.gz → 2.23.0__tar.gz

{voice_vibecoder-2.22.0 → voice_vibecoder-2.23.0}/CHANGELOG.md

@@ -2,6 +2,14 @@
 
 <!-- version list -->
 
+## v2.23.0 (2026-06-30)
+
+### Features
+
+- **audio**: Full-duplex barge-in (interrupt the model mid-response)
+  ([`0a8d90d`](https://github.com/snokam/voice-vibecoder/commit/0a8d90db88314e4fcdf7aea9bb05d7c393cfc68d))
+
+
 ## v2.22.0 (2026-06-30)
 
 ### Features

{voice_vibecoder-2.22.0 → voice_vibecoder-2.23.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: voice-vibecoder
-Version: 2.22.0
+Version: 2.23.0
 Summary: Olaf The Vibecoder — a voice-controlled coding assistant using OpenAI Realtime API + Claude Code
 Project-URL: Homepage, https://github.com/snokam/voice-vibecoder
 Project-URL: Repository, https://github.com/snokam/voice-vibecoder

{voice_vibecoder-2.22.0 → voice_vibecoder-2.23.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "voice-vibecoder"
-version = "2.22.0"
+version = "2.23.0"
 description = "Olaf The Vibecoder — a voice-controlled coding assistant using OpenAI Realtime API + Claude Code"
 readme = "README.md"
 requires-python = ">=3.10"

{voice_vibecoder-2.22.0 → voice_vibecoder-2.23.0}/src/voice_vibecoder/__init__.py

@@ -3,7 +3,7 @@
 A voice-controlled coding assistant using OpenAI Realtime API + Claude Code.
 """
 
-__version__ = "2.22.0"
+__version__ = "2.23.0"
 
 from voice_vibecoder.app_config import VoiceConfig
 from voice_vibecoder.ui.screen import VoiceCodingScreen

{voice_vibecoder-2.22.0 → voice_vibecoder-2.23.0}/src/voice_vibecoder/audio.py

@@ -20,12 +20,23 @@ class AudioManager:
     """Manages mic input and speaker output for the Realtime API.
 
     Audio format: PCM16, 24 kHz, mono — matches OpenAI Realtime requirements.
-    Auto-suppresses mic input while playing back audio to prevent
-    speaker-to-mic feedback from triggering the server VAD.
+
+    `barge_in` (default True) keeps the mic open during playback so the server
+    VAD can hear the user and interrupt the response mid-sentence — the natural
+    voice-assistant behaviour. The trade-off is that it requires headphones: on
+    open speakers the model's own audio echoes back into the mic and it will
+    interrupt itself. Set `barge_in=False` for the old half-duplex mode that
+    suppresses the mic while audio is playing (no echo, but no barge-in).
     """
 
-    def __init__(self, on_audio_chunk: Callable[[str], None]) -> None:
+    def __init__(
+        self,
+        on_audio_chunk: Callable[[str], None],
+        *,
+        barge_in: bool = True,
+    ) -> None:
         self._on_audio_chunk = on_audio_chunk
+        self._barge_in = barge_in
         self._input_stream: sd.InputStream | None = None
         self._output_stream: sd.OutputStream | None = None
         self._playback_queue: queue.Queue[bytes] = queue.Queue()
@@ -88,7 +99,9 @@ class AudioManager:
                 break
 
     def _mic_callback(self, indata: np.ndarray, frames: int, time_info, status) -> None:
-        if self._muted or self._playing or not self._running:
+        # In half-duplex mode the mic is suppressed while audio plays; in
+        # barge-in mode it stays open so the server VAD can interrupt playback.
+        if self._muted or not self._running or (self._playing and not self._barge_in):
             return
         pcm_bytes = indata.tobytes()
         encoded = base64.b64encode(pcm_bytes).decode("ascii")