Speech API

Kazakh, Russian and English: text to speech, speech to text and voice cloning over plain HTTPS.

Getting started

Sign in with your phone number — the code arrives in WhatsApp.
Top up your balance and create an API key (zs_live_…). It is shown once.
Send it as Authorization: Bearer zs_live_… to https://www.zerdesoft.kz/v1.

Base URL	`https://www.zerdesoft.kz/v1`
Text to speech	3.50 ₸ per 1,000 characters
Speech to text	3.50 ₸ per minute of audio (min. 15 s per request)
Voice cloning	10 ₸ per request + character price

Every billed response carries X-Request-Id, X-Cost-KZT and X-Balance-KZT. Requests that fail on our side are refunded automatically.

Text to speech POST /v1/audio/speech

input	required	Text, up to 10,000 characters. Numbers, dates, currency and abbreviations are read out correctly.
voice		Voice id (see voices). Unknown ids fall back to the default.
language		`kk`, `ru`, `en`. Detected automatically when omitted.
speed		0.5 – 2.0, default 0.9

Returns audio/wav (24 kHz, 16-bit, mono).

curl https://www.zerdesoft.kz/v1/audio/speech \
  -H "Authorization: Bearer $ZERDE_KEY" \
  -H "Content-Type: application/json" \
  -d '{"input": "Сәлеметсіз бе!", "voice": "quat"}' \
  -o hello.wav

Speech to text POST /v1/audio/transcriptions

file	required	mp3, wav, m4a, ogg, flac, webm, mp4 … up to 200 MB
language		`kk` (default), `ru`, `en`
response_format		`json` (default), `verbose_json` (segments with timings), `text`, `srt`, `vtt`
script		Kazakh output script: `cyrl` (default), `latyn`, `tote`
punctuate		`true` (default) restores punctuation for Kazakh

import requests

r = requests.post(
    "https://www.zerdesoft.kz/v1/audio/transcriptions",
    headers={"Authorization": f"Bearer {KEY}"},   # let requests set Content-Type
    files={"file": open("meeting.mp3", "rb")},
    data={"language": "kk", "response_format": "srt"},
    timeout=900,
)
r.raise_for_status()
print(r.text)

Voice cloning POST /v1/audio/clone

ref_audio	required	5–15 s of clean speech by the target speaker, up to 20 MB
text	required	What the cloned voice should say, up to 2,000 characters
ref_text		Exact transcript of `ref_audio`. Recognised automatically when omitted.
language		`kk` (default), `ru`, `en`

Only clone voices you own or have the speaker’s permission to use.

curl https://www.zerdesoft.kz/v1/audio/clone \
  -H "Authorization: Bearer $ZERDE_KEY" \
  -F ref_audio=@my_voice.m4a -F text="Бұл менің цифрлық дауысым." -o cloned.wav

Voices GET /v1/voices

quat	Қуат	男 · default
murat_short	Мұрат	男
girl	Аружан	女

Balance GET /v1/balance

{"balance": "10950.25", "currency": "KZT",
 "prices": {"tts_per_1k_chars": "3.50", "asr_per_minute": "3.50", "asr_min_seconds": 15, "clone_per_request": "10"}}

Errors

All errors are JSON: {"error": {"code": "…"}}. Branch on code.

400	`empty_text`, `text_too_long`, `file_required`, `bad_audio`, `reference_too_short` — fix the request
401	`invalid_api_key` — key missing, wrong or deleted
402	`insufficient_balance` — top up; `balance` and `required` are included. Nothing was charged.
413	`file_too_large`
422	`reference_not_understood` — no speech in the clone reference
429	`rate_limited` — see `Retry-After`. Default: 120 requests per minute per key.
502/503	`service_unavailable`, `busy` — temporary; retry with backoff. You were not charged.