scaletempo

Scale tempo while maintaining pitch (WSOLA-like technique with cross correlation) Inspired by SoundTouch library by Olli Parviainen

Use Sceletempo to apply playback rates without the chipmunk effect.

Example pipelines

 filesrc location=media.ext ! decodebin name=d \
     d. ! queue ! audioconvert ! audioresample ! scaletempo ! audioconvert ! audioresample ! autoaudiosink \
     d. ! queue ! videoconvert ! autovideosink

OR

 playbin uri=... audio_sink="scaletempo ! audioconvert ! audioresample ! autoaudiosink"

When an application sends a seek event with rate != 1.0, Scaletempo applies the rate change by scaling the tempo without scaling the pitch.

Scaletempo works by producing audio in constant sized chunks (#GstScaletempo:stride) but consuming chunks proportional to the playback rate.

Scaletempo then smooths the output by blending the end of one stride with the next (#GstScaletempo:overlap).

Scaletempo smooths the overlap further by searching within the input buffer for the best overlap position. Scaletempo uses a statistical cross correlation (roughly a dot-product). Scaletempo consumes most of its CPU cycles here. One can use the search propery to tune how far the algorithm looks.

Scaletempo also supports an alternative mode where a scaling factor is dynamically selected to scale input data down to the duration of the input buffers.

The use case for this is when text to speech / speech synthesis elements are placed upstream: they will attach the duration of the input text as a custom GstScaletempoTargetDurationMeta to the audio buffers they output, scaletempo can then rescale the audio down to the expected duration.

When this mode is selected, using a rate != 1.0 is not supported.