gandr-haystack

Gandr text to speech as a Haystack component.

Install

pip install gandr-haystack

Use

import os
from gandr_haystack import GandrTTS

tts = GandrTTS(voice="gandr-mia")          # reads GANDR_API_KEY from the env
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

pipe = Pipeline()
pipe.add_component("tts", GandrTTS())
pipe.run({"tts": {"text": "Hello from Haystack."}})

Parameters

run() returns {"audio": bytes, "path": str | None}. Pass path to write the WAV to disk as well as returning it.

The component serializes with to_dict() and from_dict(), and the key is held as a Haystack Secret read from the environment, so saved pipelines never carry the token.

Loading a saved pipeline

Haystack will not deserialize a third-party component unless its module is allowlisted. That is Haystack's security model, not a quirk of this package, and it applies to every community integration:

from haystack.core.serialization import allow_deserialization_module

allow_deserialization_module("gandr_haystack.tts")

Or set HAYSTACK_DESERIALIZATION_ALLOWLIST=gandr_haystack.tts in the environment. Saving with to_dict() needs nothing.

Errors

An HTTP failure raises RuntimeError carrying the API's own status and message, so a 401 (key) and a 402 (quota) are told apart rather than guessed at.

Pricing

$10 per million tokens, or $150 a month for one unlimited concurrent stream. Free key, 50,000 tokens: gandr.ai

License

MIT.