Speech

Helmcode covers both directions of speech through OpenAI-compatible audio endpoints — text-to-speech with kokoro, speech-to-text with whisper-large-v3.

Text to speech

kokoro synthesizes natural speech with sub-second latency and 67 voices (Spanish included). Call /v1/audio/speech.

Python

from openai import OpenAI
client = OpenAI(base_url="https://api.helmcode.com/v1", api_key="sk-your-key-here")

audio = client.audio.speech.create(
    model="kokoro",
    voice="ef_dora",          # Spanish female — see voices below
    input="Hola, esto es Helmcode.",
)
audio.stream_to_file("hello.mp3")

Voices include ef_dora (ES female), em_alex (ES male) and af_heart (EN female). Limit: 15 rpm.

Speech to text

whisper-large-v3 transcribes 99+ languages with automatic detection (~3.2% WER in Spanish). Call /v1/audio/transcriptions.

Python

transcript = client.audio.transcriptions.create(
    model="whisper-large-v3",
    file=open("meeting.mp3", "rb"),
)
print(transcript.text)

cURL

curl https://api.helmcode.com/v1/audio/transcriptions \
  -H "Authorization: Bearer sk-your-key-here" \
  -F model="whisper-large-v3" \
  -F file=@meeting.mp3

Files are limited to 25MB / ~2 minutes per request, at 10 rpm — split long audio into chunks. OGG/Opus and MP3 compress best.