Text becomes voice. Voice becomes text.

curl -X POST https://klangzeile.com/v1/tts \
  -H "Authorization: Bearer sk_live_…" \
  -H "Idempotency-Key: order-4711-confirmed" \
  -H "Content-Type: application/json" \
  -d '{"text": "Your order is on its way.", "voice": "en-heart", "format": "mp3"}' \
  -o confirmation.mp3

# response headers
Content-Type: audio/mpeg
X-Characters-Billed: 25

import httpx

r = httpx.post("https://klangzeile.com/v1/tts",
    headers={"Authorization": "Bearer sk_live_…",
             "Idempotency-Key": "order-4711-confirmed"},
    json={"text": "Your order is on its way.",
          "voice": "en-heart", "format": "mp3"})
r.raise_for_status()
open("confirmation.mp3", "wb").write(r.content)
print(r.headers["X-Characters-Billed"])  # 25

const r = await fetch("https://klangzeile.com/v1/tts", {
  method: "POST",
  headers: { Authorization: "Bearer sk_live_…",
             "Idempotency-Key": "order-4711-confirmed",
             "Content-Type": "application/json" },
  body: JSON.stringify({ text: "Your order is on its way.",
                         voice: "en-heart", format: "mp3" })
});
if (!r.ok) throw await r.json();   // RFC 7807 problem+json, with request_id
const audio = new Audio(URL.createObjectURL(await r.blob()));
audio.play();

Listen, transcribe, summarize.

Upload a recording — a meeting, an interview, a voice memo — and get the transcript back with timestamps, speaker labels on request, plus a structured summary: title, abstract, outline, decisions, open points. German and English. Up to three hours per file; an hour of audio is back in about two minutes. Word timestamps and SRT/VTT subtitles on request. Translation de↔en on request.

copycurl -X POST https://klangzeile.com/v1/transcriptions \
  -H "Authorization: Bearer $KEY" \
  -F "file=@meeting.mp3" -F "language=de" -F "summary=true" \
  -F "speakers=true" -F "vocabulary=Mädchenflohmarkt,Flying Circus TV"
# → 202 {"id": "tr_…", "status": "queued", "duration_seconds": 1834, …}

curl "https://klangzeile.com/v1/transcriptions/tr_…?speaker_names=S1:Host,S2:Guest" \
  -H "Authorization: Bearer $KEY"
# → 200 {"status": "done", "text": "…",
#        "segments": [{"start": 0.0, "end": 4.2, "text": "…", "speaker": "Host"}],
#        "dialogue": "Host: …\nGuest: …", "speaker_count": 2,
#        "summary": {"title": "…", "abstract": "…", "outline": [], "decisions": [], "open_points": []},
#        "minutes_billed": 31}

Poll GET or pass webhook_url and we call you when it is done — the webhook is a notification only: it carries the job id and status, never the transcript, and it is unauthenticated, so always fetch the result with GET /v1/transcriptions/{id} and your API key (signature verification is a documented follow-up). The result is delivered once: the first successful GET deletes it from our side; unread results expire after 24 hours. Billing is per started minute of audio, at the moment the job finishes — a failed job costs nothing. Speech-to-text, speaker labels and the summary are produced by Mistral AI in the EU under a zero-data-retention agreement: audio and transcript are used to answer your request and are not stored there. Speakers come back as S1, S2, …; GET …/speakers shows who said what first, then fetch the result with speaker_names=S1:Anna to have the labels replaced — we never guess a name.

• Deleted after one fetch.Text in, audio out — nothing remains but a count and a SHA-256 of the request. Uploaded recordings live in memory only — never on disk — until they are handed to the transcription engine; the transcript is stored encrypted only until you fetch it (24 h at most), then both are gone.
• No voice cloning.Five fixed voices under permissive licences. We will not build a "sound like anyone" feature.
• Hosted in the EU.Servers in Nuremberg. Voices run on our own infrastructure; speech-to-text and summaries run at Mistral AI in the EU under zero data retention — audio and text are processed for your request and not kept.
• Pinned engines.Kokoro 82M v1.0 and Piper are version-pinned by hash. The same text gives the same audio until we say otherwise.
• Limits, not surprises.When the free monthly allowance and any credit run out the API answers 402. Nothing beyond what you have is ever billed.
• Boring, on purpose.Bearer keys, JSON, standard headers. Nothing to install, nothing to learn twice.

Need more than the free allowance, or an invoice instead of a card? Write to us.