[LLM/TTS 与对话闭环]：完成语音对话主链路，包含上下文管理、NewAPI 兼容 LLM、分句 TTS、状态机和错误恢复测试

2026-06-17 18:21:56 +08:00
parent 4d6232ed29
commit e544a60eb3
7 changed files with 636 additions and 6 deletions
@@ -0,0 +1,123 @@
+from __future__ import annotations
+
+import math
+import struct
+import subprocess
+import tempfile
+from pathlib import Path
+
+from .models import AudioSegment, ErrorCode, ProviderError
+
+
+class SentenceBuffer:
+    def __init__(self, max_chars: int = 80) -> None:
+        self.max_chars = max_chars
+        self._buffer = ""
+
+    def feed(self, text: str, force: bool = False) -> list[str]:
+        if text:
+            self._buffer += text
+        chunks: list[str] = []
+        while self._should_emit(force):
+            chunks.append(self._pop_chunk(force))
+            force = False
+        return [chunk for chunk in chunks if chunk.strip()]
+
+    def flush(self) -> list[str]:
+        return self.feed("", force=True)
+
+    def _should_emit(self, force: bool) -> bool:
+        stripped = self._buffer.strip()
+        if not stripped:
+            return False
+        return force or stripped.endswith(("。", "！", "？", ".", "!", "?", "\n")) or len(stripped) >= self.max_chars
+
+    def _pop_chunk(self, force: bool) -> str:
+        stripped = self._buffer.strip()
+        if force or len(stripped) <= self.max_chars:
+            self._buffer = ""
+            return stripped
+        chunk = stripped[: self.max_chars]
+        self._buffer = stripped[self.max_chars :]
+        return chunk
+
+
+class SineTtsProvider:
+    def __init__(self, sample_rate: int = 16000) -> None:
+        self.sample_rate = sample_rate
+        self.loaded = False
+
+    def load(self) -> None:
+        self.loaded = True
+
+    def synthesize(self, text: str) -> AudioSegment:
+        if not self.loaded:
+            raise ProviderError(
+                ErrorCode.TTS_SYNTHESIS_FAILED,
+                "TTS provider is not loaded",
+                False,
+                "sine-tts",
+                "tts",
+            )
+        clean = text.strip()
+        if not clean:
+            raise ProviderError(
+                ErrorCode.TTS_EMPTY_AUDIO,
+                "cannot synthesize empty text",
+                True,
+                "sine-tts",
+                "tts",
+            )
+        duration_ms = max(120, min(1200, len(clean) * 45))
+        samples = int(self.sample_rate * duration_ms / 1000)
+        pcm = bytearray()
+        for idx in range(samples):
+            sample = int(math.sin(2 * math.pi * 440 * idx / self.sample_rate) * 12000)
+            pcm.extend(struct.pack("<h", sample))
+        return AudioSegment(bytes(pcm), self.sample_rate, 1, 0, duration_ms, {"text": clean})
+
+
+class MacSayTtsProvider:
+    def __init__(self, voice: str | None = None) -> None:
+        self.voice = voice
+        self.loaded = False
+
+    def load(self) -> None:
+        self.loaded = True
+
+    def synthesize(self, text: str) -> AudioSegment:
+        clean = text.strip()
+        if not clean:
+            raise ProviderError(
+                ErrorCode.TTS_EMPTY_AUDIO,
+                "cannot synthesize empty text",
+                True,
+                "macos-say",
+                "tts",
+            )
+        with tempfile.TemporaryDirectory() as tmp:
+            output = Path(tmp) / "speech.aiff"
+            command = ["say", "-o", str(output)]
+            if self.voice:
+                command.extend(["-v", self.voice])
+            command.append(clean)
+            try:
+                subprocess.run(command, check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+            except (FileNotFoundError, subprocess.CalledProcessError) as exc:
+                raise ProviderError(
+                    ErrorCode.TTS_SYNTHESIS_FAILED,
+                    f"macOS say failed: {exc}",
+                    False,
+                    "macos-say",
+                    "tts",
+                ) from exc
+            data = output.read_bytes()
+        if not data:
+            raise ProviderError(
+                ErrorCode.TTS_EMPTY_AUDIO,
+                "macOS say produced empty audio",
+                True,
+                "macos-say",
+                "tts",
+            )
+        return AudioSegment(data, 16000, 1, 0, max(120, len(clean) * 45), {"text": clean, "format": "aiff"})