Struct AudioChunk 

pub struct AudioChunk { /* private fields */ }

One exported block of decoded audio: owned, interleaved 32-bit float samples plus metadata.

Samples are packed native-endian f32, interleaved: as_slice.len() == frames * channels, channels interleaved ([L0, R0, L1, R1, …] for stereo) — the buffer layout whisper / candle / ort pipelines consume, at whatever sample rate and channel shape this chunk reports (normalize via the extractor’s sample_rate/channels options when a model needs a fixed shape). One chunk corresponds to one filtered AVFrame; the number of frames per chunk is not contractual (typically ~1024) and must not be relied upon.

Implementations

impl AudioChunk

pub fn pts_us(&self) -> Option<i64>

Presentation time in microseconds, passed through from the source frame and normalized to the start of the extraction window (the stream start when no start_time_us was set). None when the frame carried no usable timestamp.

pub fn index(&self) -> u64

0-based export index (counts delivered chunks in order).

pub fn sample_rate(&self) -> u32

Samples per second.

pub fn channels(&self) -> u16

Number of interleaved channels (1 for mono, 2 for stereo).

pub fn channel_layout(&self) -> &str

FFmpeg’s textual channel-layout description of this chunk (e.g. "mono", "stereo", "5.1"), read from the exported frame: the source layout under the default passthrough, or the converted layout when channels requested one. Above two channels this distinguishes layouts a bare count cannot (6 channels may be "5.1" or "6.0"). Empty when FFmpeg cannot describe the layout.

pub fn as_slice(&self) -> &[f32]

The interleaved f32 samples. Length is frames * channels.

pub fn into_vec(self) -> Vec<f32>

Consumes the chunk and returns the owned interleaved buffer (no copy).

Trait Implementations

impl Debug for AudioChunk

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.