[Feat] Streaming TTS word-level timestamps via forced aligner (#4034)

Signed-off-by: wjinxu <1299461899@qq.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
authored and GitHub committed 2026-06-22 04:51:45 +00:00
1 parent a60242092a
commit ad07f98b73
17 files changed
+2162 -20

No files matched your search

@@ -145,6 +145,15 @@ class OpenAICreateSpeechRequest(BaseModel):
default=None,
description=("Optional model-specific parameters passed directly to the model's extra_args."),
)
word_timestamps: bool = Field(
default=False,
description=(
"When true, the server runs a shared forced aligner alongside the streamed "
"audio and emits per-chunk word timestamps. Requires the server to be "
"launched with --forced-aligner pointing at an aligner model. No effect "
"when streaming is off."
),
)
@field_validator("stream_format")
@classmethod
@@ -463,6 +472,15 @@ class StreamingSpeechSessionConfig(BaseModel):
"'clause' also splits on CJK commas , and semicolons ;."
),
)
word_timestamps: bool = Field(
default=False,
description=(
"When true, audio chunks are wrapped in JSON 'audio.chunk' frames carrying "
"base64-encoded PCM plus aligned word timestamps. Requires the server to be "
"launched with --forced-aligner. When false, audio is sent as raw binary "
"frames (existing behavior)."
),
)
@model_validator(mode="after")
def validate_streaming_constraints(self) -> "StreamingSpeechSessionConfig":