Struct OnlineStream 

pub struct OnlineStream { /* private fields */ }

Context state for streaming speech recognition.

You can do VAD if you want to reduce compute utilization, but feeding constant streaming audio into this is perfectly reasonable. Decoding is incremental and constant latency.

Created by Model::online_stream.

Implementations

impl OnlineStream

pub unsafe fn flush_buffers(&mut self)

Flush extant buffers (feature frames) and signal that no further inputs will be made available.

Safety

Do not call OnlineStream::accept_waveform after calling this function.

That restriction makes it quite useless, so ymmv. I have not observed any problems doing so so long as an intervening call to OnlineStream::reset exists:

unsafe { s.flush_buffers() };
s.decode();
s.reset();
s.accept_waveform(16000, &[ ... ]);

Regardless, upstream docs state not to call OnlineStream::accept_waveform after, so do so at your own risk.

pub fn accept_waveform(&mut self, sample_rate: usize, samples: &[f32])

Accept ((-1, 1)) normalized) input audio samples and buffer the computed feature frames.

pub fn is_ready(&self) -> bool

Returns true if there are enough feature frames for decoding.

pub fn decode(&mut self)

Decode all available feature frames.

pub unsafe fn decode_unchecked(&mut self)

Decode an unspecified number of feature frames.

It is a logic error to call this function when OnlineStream::is_ready returns false.

Safety

Ensure OnlineStream::is_ready returns true. It is probably not ever worth eliding the check, but hey, you do you.

pub fn decode_batch<I: IntoIterator<Item = Q>, Q: DerefMut<Target = Self>>( streams: I, )

Decode all available feature frames in the provided iterator of streams concurrently.

This batches all operations together, and thus is superior to calling OnlineStream::decode on every OnlineStream in separate threads (though it is not invalid to do so, if desired).

pub fn decode_shared(&mut self)

Decode all available feature frames in a shared concurrency context.

This introduces a small amount of synchronization overhead in exchange for much better compute utilization.

pub fn result(&self) -> Result<String>

Returns recognition state since the last call to OnlineStream::reset.

pub fn result_with<F: FnOnce(Cow<'_, str>) -> R, R>(&self, f: F) -> Result<R>

Returns recognition state since the last call to OnlineStream::reset.

pub fn is_endpoint(&self) -> bool

Returns true if an endpoint has been detected.

pub fn reset(&mut self)

Clear any extant neural network and decoder states.

pub fn sample_rate(&self) -> usize

Returns the native sample rate.

pub fn chunk_size(&self) -> usize

Returns the chunk size at the native sample rate.

The stream becomes ready for decoding once this many samples have been accepted.

Trait Implementations

impl Drop for OnlineStream

fn drop(&mut self)

Executes the destructor for this type.

impl Send for OnlineStream

impl Sync for OnlineStream