openspeechapi 0.2.0__tar.gz → 0.2.1__tar.gz

{openspeechapi-0.2.0 → openspeechapi-0.2.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: openspeechapi
-Version: 0.2.0
+Version: 0.2.1
 Summary: Unified speech interface for STT/TTS providers
 Requires-Python: >=3.11
 Requires-Dist: httpx>=0.27

{openspeechapi-0.2.0 → openspeechapi-0.2.1}/openspeechapi/providers/tts/iflytek.py

@@ -28,6 +28,28 @@ class IflytekTTSSettings(BaseSettings):
     api_secret: str = ""
     voice: str = "xiaoyan"
     speed: int = 50
+    # Audio output encoding requested from iFlytek.
+    #   - "lame": MP3 frames (default; smaller, but caller must decode)
+    #   - "raw":  16-bit PCM @ 16 kHz mono, big-endian L16 (drop-in
+    #             playable as raw PCM; required by callers that wrap the
+    #             bytes in a fixed-format wire envelope and assume PCM,
+    #             e.g. wallex's RESP_VOICE which advertises
+    #             encoding=raw/bitDepth=16/sampleRate=16000 to the
+    #             panel — feeding MP3 bytes through that envelope plays
+    #             back as pure noise on the speaker).
+    # Default stays "lame" for backward-compat; deployments that need
+    # PCM (wallex / direct hardware playback) override via yaml.
+    aue: str = "lame"
+    # Output sample rate for raw PCM mode (only meaningful when
+    # aue="raw"). 16000 matches what the panel and the iFlytek
+    # default both expect; left configurable so an 8 kHz device or
+    # a 24 kHz studio path doesn't have to fork the provider.
+    auf_rate: int = 16000
+    # Volume (0-100). iFlytek default is 50; making it explicit so
+    # the parameter parity with Java's TtsConfig is complete.
+    volume: int = 50
+    # Pitch (0-100). Same rationale as volume.
+    pitch: int = 50
 
 class IflytekTTS(TTSProvider):
     name = "iflytek-tts"
@@ -35,12 +57,17 @@ class IflytekTTS(TTSProvider):
     execution_mode = ExecMode.IN_PROCESS
     settings_cls = IflytekTTSSettings
     capabilities = {Capability.BATCH, Capability.STREAMING, Capability.MULTILINGUAL}
-    field_options = {"voice": [
-        "xiaoyan", "aisjiuxu", "aisxping", "aisjinger", "aisbabyxu",
-        "x4_lingxiaolu_em", "x4_lingfeizhe_em", "xiaoyu", "xiaoqi",
-        "xiaofeng", "xiaomei", "xiaolin", "xiaorong", "xiaoqian",
-        "catherine", "john", "laura", "yuka", "xiaoqiukor",
-    ]}
+    field_options = {
+        "voice": [
+            "xiaoyan", "aisjiuxu", "aisxping", "aisjinger", "aisbabyxu",
+            "x4_lingxiaolu_em", "x4_lingfeizhe_em", "xiaoyu", "xiaoqi",
+            "xiaofeng", "xiaomei", "xiaolin", "xiaorong", "xiaoqian",
+            "catherine", "john", "laura", "yuka", "xiaoqiukor",
+            # English assistant-style voices used by wallex deployments.
+            "x4_enuk_ashleigh_assist",
+        ],
+        "aue": ["lame", "raw"],
+    }
 
     _WS_HOST = "tts-api.xfyun.cn"
     _WS_PATH = "/v2/tts"
@@ -107,12 +134,21 @@ class IflytekTTS(TTSProvider):
         async for chunk in self.synthesize_stream(text, opts):
             parts.append(chunk.data)
         audio_bytes = b"".join(parts)
-        logger.info("iFlytek TTS: {} chunks, {} MP3 bytes total", len(parts), len(audio_bytes))
+        # Format / sample_rate must reflect what iFlytek actually returned —
+        # callers downstream may set wire-protocol encoding metadata from
+        # this field, and a wrong tag on the bytes plays back as noise on
+        # raw-PCM consumers.
+        fmt = "pcm_s16le" if self.settings.aue == "raw" else "mp3"
+        sr = self.settings.auf_rate if self.settings.aue == "raw" else 16000
+        logger.info(
+            "iFlytek TTS: {} chunks, {} bytes total, format={}, sample_rate={}",
+            len(parts), len(audio_bytes), fmt, sr,
+        )
         return AudioData(
             data=audio_bytes,
-            sample_rate=16000,
+            sample_rate=sr,
             channels=1,
-            format="mp3",
+            format=fmt,
         )
 
     # iFlytek voice catalog — vcn → (display name, language, group)
@@ -147,15 +183,30 @@ class IflytekTTS(TTSProvider):
         vcn = voice or self.settings.voice
         spd = int(speed * 50) if speed and speed != 1.0 else self.settings.speed
         text_b64 = base64.b64encode(text.encode("utf-8")).decode("utf-8")
+        business: dict[str, Any] = {
+            "aue": self.settings.aue,
+            "vcn": vcn,
+            "speed": spd,
+            "volume": self.settings.volume,
+            "pitch": self.settings.pitch,
+            "tte": "UTF8",
+        }
+        if self.settings.aue == "lame":
+            # ``sfl=1`` (stream-frame-length) is an MP3-only knob that
+            # tells iFlytek to emit per-frame audio rather than waiting
+            # for the whole file. It has no meaning for raw PCM (raw is
+            # always streamed as 40 ms slices) and iFlytek rejects the
+            # combo with a code 10005 "invalid parameter" — so we only
+            # send it on the lame path.
+            business["sfl"] = 1
+        else:
+            # Raw / L16 mode requires ``auf`` to declare the PCM
+            # sample-rate iFlytek should produce. Java wallex sends
+            # ``audio/L16;rate=16000`` here; we mirror that exactly.
+            business["auf"] = f"audio/L16;rate={self.settings.auf_rate}"
         return {
             "common": {"app_id": self.settings.app_id},
-            "business": {
-                "aue": "lame",
-                "sfl": 1,
-                "vcn": vcn,
-                "speed": spd,
-                "tte": "UTF8",
-            },
+            "business": business,
             "data": {
                 "status": 2,
                 "text": text_b64,

{openspeechapi-0.2.0 → openspeechapi-0.2.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "openspeechapi"
-version = "0.2.0"
+version = "0.2.1"
 description = "Unified speech interface for STT/TTS providers"
 requires-python = ">=3.11"
 dependencies = [