Aud_Intf API User Guide

[中文]

Important

The audio components used in AIDK have functional differences compared to other bk_avdk branches. New features have been added specifically for AI conversation scenarios, including: - G.722 codec support - Prompt tone playback support (PCM, WAV, MP3 formats) - ASR offline voice wake-up functionality - Speaker playback source switching capability

This document provides detailed explanations of AIDK’s new features. For standard recording, playback, and voice call functionalities, please refer to:

https://docs.bekencorp.com/arminodoc/bk_avdk/bk7258/en/v2.0.1/api-reference/multi_media/bk_aud_intf.html

1. G.722 Codec

Enable G.722 codec by setting the macro CONFIG_AUD_INTF_SUPPORT_G722=y on CPU1.

2. Speaker Playback Source Switching

For scenarios requiring prompt tone interruptions during AI voice conversations, AIDK’s audio components support dynamic playback source switching via API.

Relevant APIs:

  • bk_err_t aud_tras_drv_voc_set_spk_source_type(spk_source_type_t type) // Set playback source

  • spk_source_type_t aud_tras_drv_get_spk_source_type(void) // Get current playback source

Important

These APIs are only supported on CPU1. Currently three playback sources are supported: - SPK_SOURCE_TYPE_VOICE: Voice call - SPK_SOURCE_TYPE_PROMPT_TONE: Prompt tone - SPK_SOURCE_TYPE_A2DP: Bluetooth music

3. Prompt Tone Playback

Source path: <source code>/bk_avdk/components/multimedia/prompt_tone_play/

As shown below: The prompt tone framework adopts a modular design with three main components: playback module (prompt_tone_play), source module (audio_source), and decoding module (audio_codec).

  1. Playback Module

    • Provides upper-layer application interface, integrating source and decoding modules to configure and control playback.

  2. Source Module

    • Abstracted as a class for extensibility, with implementations for different sources.

    • Current implementations: array-based (audio_array.c) and VFS filesystem (audio_vfs.c).

    • Default uses SD NAND with VFS. Select implementation via macros.

  3. Decoding Module

    • Abstracted as a class with format-specific implementations.

    • Current implementations: PCM (pcm_codec.c), WAV (wav_codec.c), MP3 (mp3_codec.c).

    • Default uses WAV format. Select via macros.

Prompt Tone Development Framework

Figure 2. Prompt Tone Development Framework

Macro configurations:

Macro

CPU

Type

Value

Description

CONFIG_AUD_INTF_SUPPORT_PROMPT_TONE

CPU1

bool

y

Enable prompt tone

CONFIG_AUD_PROMPT_TONE_SOURCE_ARRAY

CPU1

bool

y

Use global array

CONFIG_AUD_PROMPT_TONE_SOURCE_VFS

CPU1

bool

y

Use SD NAND

CONFIG_PROMPT_TONE_CODEC_PCM

CPU1

bool

y

PCM format

CONFIG_PROMPT_TONE_CODEC_WAV

CPU1

bool

y

WAV format

CONFIG_PROMPT_TONE_CODEC_MP3

CPU1

bool

y

MP3 format

Important

Prompt tone functionality depends on speaker source switching. Ensure this feature is enabled when using prompt tones.

Note

Prompt tone files must meet:

  • Mono channel

  • 16-bit depth

  • 16kHz sample rate

Important

  • The WAV format prompt tone file must strictly follow the format of 44-byte header + valid PCM audio data and must not contain additional information such as author or album details.

  • The MP3 format prompt tone file must not include an ID3v1 header or other metadata such as author or album information.

4. ASR Offline Voice Wake-up

AIDK supports ASR offline wake-up for AI conversations. The ASR algorithm runs on CPU2 (enable via CONFIG_AI_ASR_MODE_CPU2=y).

When enabled, the device requires a wake word before processing microphone data for AI Agent.