Nabrah TTS V1

Nabrah_tts_v1

Stable

Nabrah TTS V1 is a text-to-speech model designed for real-time voice applications. It converts text into natural speech and supports both streaming and non-streaming generation.

The model can be used with Nabrah's available voices as well as custom cloned voices.

Model overview

Recommended use cases

Nabrah TTS V1 is suitable for applications that require generated speech, including:

• Voice agents

• Customer service applications

• IVR systems

• Conversational AI

• Interactive voice experiences

• Content narration

• Real-time speech generation

Streaming and non-streaming

Nabrah TTS V1 supports both streaming and non-streaming speech generation.

Streaming

In streaming mode, audio is delivered progressively as it is generated. This allows applications to begin receiving audio before the full speech output has been synthesized.

Streaming is recommended for latency-sensitive applications such as voice agents and real-time conversational systems.

Server-side TTFB: 120 –130 ms

TTFB represents the server-side time until the first response bytes are returned. It does not represent full end-to-end application latency.

Non-streaming

In non-streaming mode, the complete audio file is returned after speech generation has finished.

This mode is suitable when progressive playback is not required, such as generating audio files for later playback or storage.

Audio output

Nabrah TTS V1 supports three output formats.

MP3

MP3 provides compressed audio at 192 kbps, making it suitable for applications where smaller payload size and broad playback compatibility are important.

WAV

WAV output uses uncompressed 16-bit PCM audio at 48 kHz mono.

It is suitable for applications that require lossless audio or a self-contained audio file with format information included in the file header.

PCM

PCM output contains raw signed 16-bit little-endian audio samples at 48 kHz mono.

It does not contain a file header and is intended for applications that consume raw audio frames directly.

Voice cloning

Nabrah TTS V1 supports custom cloned voices.

A cloned voice can be used for speech generation after the cloning process has completed and the voice is marked as ready.

The same voice IDs can be used when configuring Nabrah voice agents.

Learn about Voice Cloning

View available Voices

Input limits

The model accepts up to:

16,384 characters per request

The limit is calculated after text normalization.

Requests that exceed the supported input length must be divided into smaller segments before synthesis.

Speech controls

Some speech control parameters are currently exposed through the API but are not yet applied to generated audio.

Support for additional speech controls may change as the model evolves.

Latency

Actual end-to-end latency may vary depending on network conditions, infrastructure, request configuration, and client-side audio handling.

Using Nabrah TTS V1

Use the following model ID when generating speech:

Nabrah_tts_v1

Speech generation is available through:

POST /api/ext/tts/generations

Example:

```json \
{ \
"model": "Nabrah_tts_v1", \
"input": "Hello from Nabrah!", \
"voice": "YOUR_VOICE_ID", \
"response_format": "mp3", \
"stream": false \
}

```

Generate Speech

API Reference