Runtime¶

Input audio stream.

This audio stream must be headerless, 16-bit LPCM-encoded, sampled at 16 kHz, with little-endian byte ordering.

push, run

Output audio stream.

Headerless audio output, 16-bit LPCM-encoded, sampled at 16 kHz, with little-endian byte ordering.

Audio stream requested start index.

Start the next audio-stream at this sample index value. Defaults to audio-stream-first.

audio-stream, audio-stream-size

Audio stream requested end index.

End the next audio-stream at this sample index value. Defaults to audio-stream-last.

audio-stream, audio-stream-size

VAD initial ignore duration, in ms.

Ignore the first skip-to-ms ms of the ->audio-pcm input stream. Use this runtime setting to skip over a trigger phrase included in the source audio. The default is to process all audio.

skip-to-sample

VAD initial ignore duration, in samples.

Ignore the first skip-to-sample samples of the ->audio-pcm input stream. Use this runtime setting to skip over a trigger phrase included in the source audio. The default is to process all audio.

skip-to-ms

Current enrollment includes trailing context.

Set to 1 if the enrollment recording should include trailing context, for example: "Hey Sensory will it rain tomorrow?"

^resume

Delete an enrolled user.

Deletes the named user, then:

• invokes ^enrolled event,
• invokes ^adapted event iff any enrolled users remain,
• invokes ^done event.

^adapted, ^done, ^enrolled

Force re-adaptation of all enrollments.

If all users in an enrollment task have been adapted, the adaptation step is skipped. This is the case when one or more adapted enrollment contexts are loaded, and no new users are added.

Setting re-adapt to 1 changes this behavior to always do the adaptation step.

^adapted

Rename an enrolled user.

Changes the recognition result returned for user to the string argument.

This setting is only available in fixed-phrase spotters that support continuous adaptation.

snsrSetString(s, SNSR_USER, "original-user/phrase");
snsrSetString(s, SNSR_RENAME_USER, "new-user/new-phrase");

session.setString(Snsr.USER, "original-user/phrase");
session.setString(Snsr.RENAME_USER, "new-user/new-phrase");

^adapted, ^new-user

Embedded device acoustic model data.

THF Micro, dsp-target, dsp-header-stream, dsp-search-stream, dsp.production-ready, dsp.t-slice-version

Embedded device search header.

dsp-target,dsp-acmodel-stream, dsp-search-stream, dsp.t-slice-version

Embedded device search model data.

dsp-target, dsp-acmodel-stream, dsp-header-stream, dsp.t-slice-version

Embedded (DSP) device target name.

THF Micro, dsp-acmodel-stream, dsp-header-stream, dsp-search-stream, dsp.t-slice-version

Model license expiration date.

Returns the license expiration date of the most recently loaded model in seconds since the epoch, or 0 if no model is loaded.

For production keys, which never expire, the expiration date is 0.

double e;
time_t expdate;
snsrGetDouble(s, SNSR_MODEL_LICENSE_EXPDATE, &e);
expdate = (time_t)e;

model-license-exp-message

Model license expiration message.

Returns an expiration message string for the most recently loaded model, or NULL if no model is loaded.

The returned string is of the form "Model license expires on <date>", or NULL for model license keys that do not expire.

const char *expires;
snsrGetString(s, SNSR_MODEL_LICENSE_EXPIRES, &expires);
if (expires) fprintf(stderr, "%s\n", expires);

model-license-exp-date, model-license-exp-warn

Model license expiration warning message.

This value is NULL for models with license keys that either do not expire, or that have an expiration date that is more than 60 days into the future.

For license keys expiring in 60 days or fewer, the returned string will be of the form "License will expire in 37 days.".

model-license-exp-date, model-license-exp-message

Source model name.

The name of the model file used to create:

• ROM-able C code with save and SOURCE.
• A model in one of the supported embedded formats with dsp-header-stream.

This value is included in the comment header of the generated code, for information only.

Prepares a Session for generating custom initialization code.

Set this value to NULL to enable load instrumentation. This is used by save with SUBSET_INIT to generate custom library initialization code (in C), which references only modules encountered during load.

Linking against only a subset of available modules reduces executable size.

This example will create a custom-init.c file. Add this to the build, and recompile with -DSNSR_USE_SUBSET to enable. The linked executable(s) will contain just the modules required to run model-a.snsr and model-b.snsr.

SnsrSession s;
snsrNew(&s);
snsrSetString(s, SNSR_PREPARE_SUBSET_INIT, NULL);
snsrLoad(s, snsrStreamFromFileName("model-a.snsr", "r"));
snsrLoad(s, snsrStreamFromFileName("model-b.snsr", "r"));
snsrSave(s, SNSR_FM_SUBSET_INIT, snsrStreamFromFileName("custom-init.c", "w"));
snsrRelease(s);

load, save, SUBSET_INIT

Prepare a Session for pruning model settings

Pruning unused settings from a model reduces peak RAM requirements. This is typically only useful on platforms where heap memory is constrained.

Set prune:enable to yes to instrument model loading and running. This instrumentation:

• Keeps track of which configuration settings the application accesses during model evaluation,
• adds a list of these settings to the model upon saving, and
• configures the model to prune settings not on this list from the runtime directly after loading.

Set prune:enable to no to disable pruning. This is the default.

load, save

Exported identifier in ROM C code.

When text segment C code is created with save and SOURCE, this setting specifies the name of the exported data structure.

The value must start with an ASCII alphabetic character or _, and contain only alphanumerics and _; it must match regular expression [A-Za-z_][A-Za-z0-9_]*