## Input requirements

| Property        | Required value                                              |
|----------------|------------------------------------------------------------|
| Channels       | Mono (1 channel)                                          |
| Sample format  | 32-bit float, normalised to −1.0 … +1.0                  |
| Sample rate    | 8 000, 16 000, or 48 000 Hz — must match `AudioParams.sampleRate` |
| Frame size     | 20 ms recommended: 160 samples @ 8 kHz, 320 @ 16 kHz, 960 @ 48 kHz |

## Real-time pacing

This is a live stream. Feed approximately one frame every 20 ms, as you would from a live microphone. Sending much faster than real time can overrun server-side buffers. In a live call scenario the microphone callback provides natural pacing; when replaying a file (as in the [Quick Start](https://developer.sanas.ai/API-Reference/Cpp-SDK/Guides/Quick-Start)), add a 20 ms sleep between frames.

## PCM conversion

```cpp
// int16 → float
float sample = pcm16 / 32768.0f;

// float → int16 (for writing WAV output)
int16_t pcm = (int16_t)(std::max(-1.0f, std::min(1.0f, sample)) * 32767.0f);
```