Integration: Gandr
Gandr Text-to-Speech component for Haystack.
Table of Contents
Overview
Gandr is a text-to-speech API. This integration adds a
GandrTTS component that turns text into WAV audio inside a Haystack pipeline,
and optionally writes it to disk.
Six stock voices, output at 8 kHz to 24 kHz, and every render is watermarked.
Installation
pip install gandr-haystack
Usage
Set GANDR_API_KEY in your environment. Keys start with gnd_, and a free key
with 100,000 characters is available at
gandr.ai without a
card.
from gandr_haystack import GandrTTS
tts = GandrTTS(voice="gandr-mia")
result = tts.run(text="Your appointment is confirmed for Tuesday at two fifteen.",
path="out.wav")
result["audio"] # WAV bytes
In a pipeline:
from haystack import Pipeline
from gandr_haystack import GandrTTS
pipe = Pipeline()
pipe.add_component("tts", GandrTTS())
pipe.run({"tts": {"text": "Hello from Haystack."}})
Parameters
| Parameter | Default | Notes |
|---|---|---|
api_key |
Secret.from_env_var("GANDR_API_KEY") |
Keys start with gnd_ |
voice |
gandr-mia |
Also gandr-ava, gandr-jenny, gandr-dane, gandr-leo, gandr-lewis |
language |
en |
ISO language code |
sample_rate |
24000 |
One of 8000, 16000, 22050, 24000 |
url |
https://tts.gandr.ai/v1/tts/bytes |
Override to pin a region |
timeout |
60.0 |
Seconds |
run() returns {"audio": bytes, "path": str | None}.
The API key is held as a Haystack Secret read from the environment, so a
serialized pipeline never carries the token. An HTTP failure raises
RuntimeError carrying the API’s own status and message, so an authentication
error and a quota error are distinguishable.
Loading a saved pipeline
Haystack does not deserialize a third-party component unless its module is allowlisted:
from haystack.core.serialization import allow_deserialization_module
allow_deserialization_module("gandr_haystack.tts")
Or set HAYSTACK_DESERIALIZATION_ALLOWLIST=gandr_haystack.tts. Saving with
to_dict() requires nothing.
License
MIT.