Provider

Free text to speech API

A text to speech (TTS) API turns text into spoken audio. Some providers give TTS free, with character or request limits. FreeLLMAPI sends /v1/audio/speech requests to the providers that serve speech models. When one provider fails or hits its limit, the router tries the next one.

Free text to speech providers

The table shows providers with a free TTS allowance. We checked each limit on the official provider page. Free speech limits use characters, requests or a daily compute budget, so compare them with care.

Free TTS allowances from provider pages, checked 20 November 2026
ProviderModelFree allowance (official)OutputNotes
GroqOrpheus English, Orpheus Arabic (Saudi)Free plan with daily limits. Maximum 200 characters for each requestWAV onlyAn organization admin must accept the model terms in the Groq console first
Google AI StudioGemini 2.5 Flash TTS (preview)Free tier: input and output are free of charge, with rate limitsWAVMany voices. Good for natural speech
Cloudflare Workers AIMeloTTS10,000 neurons each day. MeloTTS uses 18.63 neurons for each audio minuteWAVAbout 536 audio minutes each day if no other model uses the budget. Language selector, no voice selector
Cloudflare Workers AIDeepgram Aura 1Same 10,000 neurons. Aura 1 uses 1,363.64 neurons for each 1,000 charactersMP3About 7,300 characters each day
SpeechifySimba500,000 characters each month, no cardMP3Commercial use is listed on the free plan. Simba 3.5 drops to 333,000 characters from 21 November 2026
CartesiaSonic20,000 credits each month. 1 credit is 1 characterWAVThe commercial use license is listed on paid plans only

The "about" numbers are our calculation from the published rates. The Cloudflare budget is shared by all Workers AI models on one account, including chat and image models.

Step 1: Install FreeLLMAPI and add speech keys

FreeLLMAPI is open source. It runs on your computer in Docker. Your provider keys stay on your computer.

Install (macOS or Linux)

curl -fsSL https://freellmapi.co/install.sh | bash

Open the dashboard at http://localhost:3001. Add your keys on the Keys page. A Groq key or a Google AI Studio key also enables their speech models. Then open Models, Audio to see the speech models and turn them on or off. Copy your unified key from the Keys page.

Step 2: Generate speech

The endpoint uses the OpenAI request format. The input field is required. The model, voice and response_format fields are optional. The response body is the audio file.

curl: let the router choose a provider

curl http://localhost:3001/v1/audio/speech \
  -H "Authorization: Bearer <unified-key>" \
  -H "Content-Type: application/json" \
  -d '{"model": "auto", "input": "Your order has shipped.", "voice": "alloy"}' \
  --dump-header headers.txt \
  --output speech.audio
grep -i -E "content-type|x-provider" headers.txt

Python: pin Groq Orpheus English

from openai import OpenAI

client = OpenAI(base_url="http://localhost:3001/v1", api_key="<unified-key>")

with client.audio.speech.with_streaming_response.create(
    model="canopylabs/orpheus-v1-english",
    voice="troy",           # a native Orpheus voice; "alloy" also maps to troy
    input="Your order has shipped. It arrives on Friday.",
    response_format="wav",  # Orpheus returns WAV only
) as response:
    print("Served by:", response.headers.get("x-provider"))
    response.stream_to_file("speech.wav")

The X-Provider header shows which provider made the audio. The Content-Type header shows the format, for example audio/wav or audio/mpeg.

Voices, formats and text length

  • Voices: standard OpenAI voice names, such as alloy, map to a tested native voice on many providers. You can also send a native voice name, for example troy for Orpheus English or fahad for Orpheus Arabic.
  • Formats: each provider makes its own format. Groq Orpheus and Cloudflare MeloTTS make WAV. Cloudflare Aura makes MP3. A provider refuses a format that it cannot make. It does not send a file with a wrong label.
  • Text length: Groq accepts at most 200 characters for each request. Split long text into sentences and send one request for each sentence.
  • Language: Orpheus has English and Saudi Arabic models. MeloTTS on Cloudflare uses English through this adapter.

Python: split long text into short requests

import re
from openai import OpenAI

client = OpenAI(base_url="http://localhost:3001/v1", api_key="<unified-key>")
text = open("article.txt").read()
parts = [s for s in re.split(r"(?<=[.!?])\s+", text) if s.strip()]

for i, part in enumerate(parts):
    with client.audio.speech.with_streaming_response.create(
        model="auto", voice="alloy", input=part[:200],
    ) as r:
        ext = "mp3" if "mpeg" in r.headers.get("content-type", "") else "wav"
        r.stream_to_file(f"part-{i:03d}.{ext}")

Rules for use and pitfalls

  • License terms: free plans can limit commercial use. Cartesia lists the commercial license on paid plans only. Read the terms of each provider before you publish audio.
  • Shared allowances: a monthly character allowance is for the account. More voices or more models do not give more characters.
  • Terms in the Groq console: Groq returns an error until an admin accepts the model terms. FreeLLMAPI does not accept terms for you.
  • Mixed formats: with model: "auto", two requests can come from different providers. Check Content-Type before you join files, or pin one model.
  • Joining parts: join the audio parts with a tool such as ffmpeg. All parts must have the same format and the same sample rate, so pin one model for one document.
  • Truncation: the split example cuts each part at 200 characters. Keep your sentences shorter than that, or a word can be lost.

Troubleshooting

ProblemCauseFix
Error 400 on a formatThe provider cannot make that formatRemove response_format, or request wav from Orpheus
Groq request fails on long textThe text has more than 200 charactersSplit the text into short parts
Groq error about termsThe model terms are not acceptedAccept the terms in the Groq console
Audio file does not playThe file extension does not match the formatUse the Content-Type header to set the extension
Error 503No speech model is enabledAdd a key and turn on a model on Models, Audio
Use every one of these through one key. FreeLLMAPI is an open-source, self-hosted router that puts all these free tiers behind a single OpenAI-compatible endpoint and fails over when one is rate-limited. Browse the catalog or go live.

Frequently asked questions

Is there a free text to speech API?

Yes. Groq, Google AI Studio, Cloudflare Workers AI, Speechify and Cartesia give free text to speech with limits. FreeLLMAPI sends /v1/audio/speech requests to these providers through one key.

Which free text to speech API has the most characters?

Of the providers in the table, Speechify lists 500,000 characters each month on its free plan. Cartesia lists 20,000 credits each month, at 1 credit for each character.

Can I use free text to speech audio commercially?

It depends on the provider. Speechify lists commercial use on its free plan. Cartesia lists the commercial license on paid plans only. Read the terms of each provider.

Does the OpenAI SDK work for free text to speech?

Yes. Set base_url to http://localhost:3001/v1 and call client.audio.speech.create. The router selects a provider and returns the audio bytes.

Sources

All guides · Browse all free models

Keep reading

All posts →
Provider

Free Gemini API

Google AI Studio's permanent free tier — Gemini and Gemma, no credit card.

Provider

Free DeepSeek API

DeepSeek V3, R1 and V4 served free across multiple providers.

Provider

Free Llama API

Meta's Llama models free across 14 providers, behind one key.