wip

livekit · Jul 23, 2024 · 0a25491 · 0a25491
1 parent baafe48
commit 0a25491
Show file tree

Hide file tree

Showing 4 changed files with 6 additions and 8 deletions.
diff --git a/livekit-agents/livekit/agents/ipc/channel.py b/livekit-agents/livekit/agents/ipc/channel.py
@@ -146,6 +146,8 @@ def _read_thread(self) -> None:
                         self._loop.call_soon_threadsafe(self._read_q.put_nowait, msg)
                     except RuntimeError:
                         break
+            except (OSError, EOFError, ValueError):
+                break
             except ChannelClosed:
                 break
 

diff --git a/livekit-agents/livekit/agents/tokenize/token_stream.py b/livekit-agents/livekit/agents/tokenize/token_stream.py
@@ -44,7 +44,7 @@ def push_text(self, text: str) -> None:
                 self._event_ch.send_nowait(TokenData(token=buf))
 
                 for i, tok in enumerate(buf_toks):
-                    tok_i = self._buf.index(tok)
+                    tok_i = self._buf.find(tok)
                     self._buf = self._buf[tok_i + len(tok) :].lstrip()
 
                 buf_toks = []

diff --git a/livekit-agents/livekit/agents/transcription/tts_forwarder.py b/livekit-agents/livekit/agents/transcription/tts_forwarder.py
@@ -175,8 +175,7 @@ def push_text(self, text: str) -> None:
     def mark_text_segment_end(self) -> None:
         self._check_not_closed()
         stream = self._forming_segments.text.sentence_stream
-        stream.mark_segment_end()
-        self._task_set.create_task(stream.aclose())
+        stream.end_input()
 
         # create a new segment on "mark_text_segment_end"
         # further text can already be pushed even if mark_audio_segment_end has not been
@@ -235,8 +234,7 @@ async def _forward_task():
             seg.forward_start_time = time.time()
 
             async for ev in sentence_stream:
-                if ev.type == tokenize.TokenEventType.TOKEN:
-                    await self._sync_sentence_co(seg, ev.token, rtc_seg_q)
+                await self._sync_sentence_co(seg, ev.token, rtc_seg_q)
 
         rtc_seg_q.put_nowait(None)
         await forward_task

diff --git a/livekit-agents/livekit/agents/voice_assistant/agent_output.py b/livekit-agents/livekit/agents/voice_assistant/agent_output.py
@@ -189,8 +189,6 @@ async def _read_generated_audio_task():
             handle._buf_ch.send_nowait(audio.frame)
 
             # we're only flushing once, so we know we can break at the end of the first segment
-            if audio.end_of_segment:
-                break
 
         # self._log_debug(
         #    f"tts finished synthesising {audio_duration:.2f}s audio (streamed)"
@@ -212,7 +210,7 @@ async def _read_generated_audio_task():
         if handle._tr_fwd is not None:
             handle._tr_fwd.mark_text_segment_end()
 
-        tts_stream.flush()
+        tts_stream.end_input()
         await read_atask
         await tts_stream.aclose()