RTF
An on-device speech inference performance metric defined as the ratio of the time required to process audio to the duration of the audio itself. RTF<1 indicates faster than real time, while RTF>1 indicates inference cannot keep pace with audio arrival.
Core Facts
Timeline (last 90 days)
In a pure CPU environment (2018 Threadripper, under load), SteerTTS edge achieves an RTF of 0.053 with 4 threads (about 19x real-time), with first-packet audio latency of about 120 ms thanks to sentence-level streaming.
WordCraft can open and save docx files, and supports ODT, RTF, HTML, Markdown, LaTeX, and plain text formats.
The median per-sentence RTF for the entire answer is essentially the same in the parallel and alternating modes (0.84 to 0.89), indicating that the bottleneck is concentrated in first-sentence synthesis rather than steady-state synthesis speed.
RTF measurement is highly dependent on preconditions such as device model, CPU cores and frequency, whether SIMD instructions are enabled, number of threads, batch processing versus streaming chunking, and audio sample rate and bit depth.
Real-time factor (RTF) refers to the ratio of the time required to process audio to the duration of the audio itself; an RTF less than 1 means processing is faster than real time; Parakeet's RTF is 0.014 and Whisper's RTF is 0.095.
An RTF (Real-Time Factor) of less than 1 means the system processes faster than the audio playback speed, which is the baseline threshold for achieving truly real-time transcription.
RTF (Real-Time Factor) is defined as the ratio of the computation time required to generate audio to the duration of that audio, with a 5x real-time speed corresponding to an RTF of 0.2.
All Facts (7)
An RTF (Real-Time Factor) of less than 1 means the system processes faster than the audio playback speed, which is the baseline threshold for achieving truly real-time transcription.
65%UnverifiedIn a pure CPU environment (2018 Threadripper, under load), SteerTTS edge achieves an RTF of 0.053 with 4 threads (about 19x real-time), with first-packet audio latency of about 120 ms thanks to sentence-level streaming.
50%UnverifiedWordCraft can open and save docx files, and supports ODT, RTF, HTML, Markdown, LaTeX, and plain text formats.
50%UnverifiedThe median per-sentence RTF for the entire answer is essentially the same in the parallel and alternating modes (0.84 to 0.89), indicating that the bottleneck is concentrated in first-sentence synthesis rather than steady-state synthesis speed.
50%UnverifiedRTF measurement is highly dependent on preconditions such as device model, CPU cores and frequency, whether SIMD instructions are enabled, number of threads, batch processing versus streaming chunking, and audio sample rate and bit depth.
50%UnverifiedReal-time factor (RTF) refers to the ratio of the time required to process audio to the duration of the audio itself; an RTF less than 1 means processing is faster than real time; Parakeet's RTF is 0.014 and Whisper's RTF is 0.095.
50%UnverifiedRTF (Real-Time Factor) is defined as the ratio of the computation time required to generate audio to the duration of that audio, with a 5x real-time speed corresponding to an RTF of 0.2.
50%