[ English | 简体中文 ]

Sound Model API

The Sound Model interface handles low-level acoustic model data required by the media trigger, providing model loading, unloading, property/option queries, and hotword detection capabilities.

Header: #include <media_trigger_model.h>

openvela Implementation Notes

• Relationship with Media Trigger: media_trigger_* is the high-level voice wakeup interface (including DSP path), while media_trigger_model_* is the low-level model operation interface (no audio capture involved)
• Typical Usage: Custom recognition flows, running one-shot keyword detection on a specific PCM buffer
• Model Properties/Options: Use get_properties to query vendor properties, and get_options to read recommended audio sampling parameters
• One-shot Detection: detect_hotword performs a single synchronous detection on a given PCM buffer

Model Lifecycle

media_trigger_model_load

void* media_trigger_model_load(const void* model, size_t size);

Loads acoustic model data and returns a model context.

Parameters:

• model Pointer to the start of model data.
• size Size of the model data in bytes.

Returns:

Returns the model context pointer on success, or NULL on failure.

media_trigger_model_unload

void media_trigger_model_unload(void* context);

Unloads a previously loaded model context.

Parameters:

• context The model context to unload.

Model Property Queries

media_trigger_model_get_properties

void media_trigger_model_get_properties(void* properties, size_t* size);

Queries vendor-provided model property information.

Parameters:

• properties Output buffer for storing property data.
• size Input/output parameter: on entry specifies the buffer size, on return updated to the actual number of bytes written.

media_trigger_model_get_options

void media_trigger_model_get_options(void* context, char* options, size_t size);

Queries the model's recommended audio capture options string, e.g. "format=s16le:sample_rate=16000:ch_layout=mono".

Parameters:

• context Model context (returned by media_trigger_model_load).
• options Output buffer for receiving the options string.
• size Output buffer size in bytes.

media_trigger_model_get_buffer_size

void media_trigger_model_get_buffer_size(void* context, size_t* size);

Queries the model's recommended recording buffer size.

Parameters:

• context Model context.
• size Output parameter, returns the recommended buffer size in bytes.

Hotword Detection

media_trigger_model_detect_hotword

bool media_trigger_model_detect_hotword(void* context, const char* buffer, size_t size);

Runs a single model detection pass on the given PCM buffer to determine whether a hotword is matched.

Parameters:

• context Model context.
• buffer PCM audio buffer to analyze.
• size Buffer size in bytes.

Returns:

Returns true if a hotword is detected, false otherwise.