GstAudioEncoder

This base class is for audio encoders turning raw audio samples into encoded audio data.

GstAudioEncoder and subclass should cooperate as follows.

Configuration

• Initially, GstAudioEncoder calls @start when the encoder element is activated, which allows subclass to perform any global setup.
• GstAudioEncoder calls @set_format to inform subclass of the format of input audio data that it is about to receive. Subclass should setup for encoding and configure various base class parameters appropriately, notably those directing desired input data handling. While unlikely, it might be called more than once, if changing input parameters require reconfiguration.
• GstAudioEncoder calls @stop at end of all processing.

As of configuration stage, and throughout processing, GstAudioEncoder maintains various parameters that provide required context, e.g. describing the format of input audio data. Conversely, subclass can and should configure these context parameters to inform base class of its expectation w.r.t. buffer handling.

Data processing

• Base class gathers input sample data (as directed by the context's frame_samples and frame_max) and provides this to subclass' @handle_frame.
• If codec processing results in encoded data, subclass should call gstaudio.audio_encoder.AudioEncoder.finishFrame to have encoded data pushed downstream. Alternatively, it might also call gstaudio.audio_encoder.AudioEncoder.finishFrame (with a NULL buffer and some number of dropped samples) to indicate dropped (non-encoded) samples.
• Just prior to actually pushing a buffer downstream, it is passed to @pre_push.
• During the parsing process GstAudioEncoderClass will handle both srcpad and sinkpad events. Sink events will be passed to subclass if @event callback has been provided.

Shutdown phase

• GstAudioEncoder class calls @stop to inform the subclass that data parsing will be stopped.

Subclass is responsible for providing pad template caps for source and sink pads. The pads need to be named "sink" and "src". It also needs to set the fixed caps on srcpad, when the format is ensured. This is typically when base class calls subclass' @set_format function, though it might be delayed until calling @gst_audio_encoder_finish_frame.

In summary, above process should have subclass concentrating on codec data processing while leaving other matters to base class, such as most notably timestamp handling. While it may exert more control in this area (see e.g. @pre_push), it is very much not recommended.

In particular, base class will either favor tracking upstream timestamps (at the possible expense of jitter) or aim to arrange for a perfect stream of output timestamps, depending on #GstAudioEncoder:perfect-timestamp. However, in the latter case, the input may not be so perfect or ideal, which is handled as follows. An input timestamp is compared with the expected timestamp as dictated by input sample stream and if the deviation is less than #GstAudioEncoder:tolerance, the deviation is discarded. Otherwise, it is considered a discontuinity and subsequent output timestamp is resynced to the new position after performing configured discontinuity processing. In the non-perfect-timestamp case, an upstream variation exceeding tolerance only leads to marking DISCONT on subsequent outgoing (while timestamps are adjusted to upstream regardless of variation). While DISCONT is also marked in the perfect-timestamp case, this one optionally (see #GstAudioEncoder:hard-resync) performs some additional steps, such as clipping of (early) input samples or draining all currently remaining input data, depending on the direction of the discontuinity.

If perfect timestamps are arranged, it is also possible to request baseclass (usually set by subclass) to provide additional buffer metadata (in OFFSET and OFFSET_END) fields according to granule defined semantics currently needed by oggmux. Specifically, OFFSET is set to granulepos (= sample count including buffer) and OFFSET_END to corresponding timestamp (as determined by same sample count and sample rate).

Things that subclass need to take care of:

• Provide pad templates
• Set source pad caps when appropriate
• Inform base class of buffer processing needs using context's frame_samples and frame_bytes.
• Set user-configurable properties to sane defaults for format and implementing codec at hand, e.g. those controlling timestamp behaviour and discontinuity processing.
• Accept data in @handle_frame and provide encoded results to gstaudio.audio_encoder.AudioEncoder.finishFrame.