Class AudioCppTiming
Inference timing. Different endpoints populate different members: non-streaming runs
report WallMs, AudioDurationMs and Rtf;
streaming runs report TtftMs (when the server sent a ttft_ms key
— check TtftPresent because a live-speech timing can carry an explicit
JSON null); live speech adds the remaining members.
public sealed class AudioCppTiming
- Inheritance
-
AudioCppTiming
- Inherited Members
Properties
AudioDurationMs
Audio duration in milliseconds (generated audio for speech, input audio for transcription).
public double? AudioDurationMs { get; }
Property Value
FirstAudioBeforeInputEnd
Live speech: whether first output audio arrived before the input stream closed.
public bool? FirstAudioBeforeInputEnd { get; }
Property Value
- bool?
InputEndMs
Live speech: milliseconds from request start to the end of the input stream.
public double? InputEndMs { get; }
Property Value
OverlapMs
Live speech: milliseconds of output that overlaps the still-open input stream.
public double? OverlapMs { get; }
Property Value
RequestStartToFirstAudioMs
Live speech: milliseconds from request start to first output audio.
public double? RequestStartToFirstAudioMs { get; }
Property Value
Rtf
Real-time factor: wall_ms / audio_duration_ms.
public double? Rtf { get; }
Property Value
TtftMs
Time to first output, in milliseconds; null when absent or explicitly null.
public double? TtftMs { get; }
Property Value
TtftPresent
true when the timing object contained a ttft_ms key with a
non-null value. A live-speech timing sets TtftMs to
null while this is true when the first output
audio arrived before the input stream ended.
public bool TtftPresent { get; }
Property Value
WallMs
Inference wall time in milliseconds.
public double? WallMs { get; }