TTS quickstart

TTS quickstart#

This page walks you from a fresh install to synthesized audio. You will write a short Python script, run it, and understand what each part does.

Prerequisites#

You need:

  • Python 3.10+ and pip or uv

  • The ABR SDK: pip install abr-sdk

  • A TTS application package downloaded and extracted from the ABR developer portal. You should have a directory containing a .so shared library file.

If you have not downloaded an application package yet, follow Installation first.

Write the script#

Create a file called tts_quickstart.py:

import subprocess
from abr_sdk.tts import Tts
from abr_sdk.tts_preprocess import tts_preprocess

LIBRARY_PATH = "/path/to/nith-5m-live.en-f-quill/libnith_5m_live.so"  # female voice

text = tts_preprocess("Hello, world. Welcome to on-device text-to-speech.").encode("utf-8")

chunks: list[bytes] = []

with Tts(LIBRARY_PATH) as tts:
    tts.push(text, on_pcm=chunks.append)
    tts.wait_for_completion()

with open("output.pcm", "wb") as f:
    f.write(b"".join(chunks))

print("Wrote output.pcm")
subprocess.run(["aplay", "-f", "S16_LE", "-r", "16000", "-c", "1", "output.pcm"])

Replace LIBRARY_PATH with the path to the shared library inside your extracted application package.

Run it#

python tts_quickstart.py

You should see Wrote output.pcm and hear the synthesized audio play automatically. To play back the audio from the generated audio file, run:

aplay -f S16_LE -r 16000 -c 1 output.pcm   # Linux

What just happened#

tts_preprocess(...).encode("utf-8"): The TTS model expects clean, speakable text with no digits or symbols. tts_preprocess converts "Hello, world." into a form the model can pronounce, then encode("utf-8") converts it to the bytes that push() expects.

with Tts(LIBRARY_PATH) as tts:: The with block opens the application package and loads the neural network on entry. On exit, whether normal or on exception, the SDK releases the loaded library and all internal buffers.

tts.push(text, on_pcm=chunks.append): Feeds the text to the model. The call returns immediately. The on_pcm callback fires as the model produces PCM audio; here it appends each chunk to the chunks list.

tts.wait_for_completion(): Signals end-of-input, flushes the synthesis pipeline, and blocks until all audio has been delivered through on_pcm.

b"".join(chunks): Concatenates the individual PCM chunks into one continuous byte string.

subprocess.run(["aplay", ...]): Plays back the saved file immediately using aplay. The -f S16_LE -r 16000 -c 1 flags tell aplay to interpret the raw bytes as signed 16-bit little-endian mono audio at 16 kHz, matching the format the TTS model produces.

Next steps

  • Input format: the full text format requirements and how to handle numbers, acronyms, and abbreviations.

  • Examples: ready-to-run examples for streaming to speakers, synthesizing multiple sentences, and reading from stdin.

  • Overview: how the streaming model works and the PCM output format in detail.