Text to SpeechPublic

Nari Qwen3-TTS 1.7B

The world’s fastest TTS endpoint. 50 ms time to first audio. Choose Fast for latency or Standard for cost efficiency.

Time to first audio

From 50 ms

Price / 1M chars

From $5

Languages

10 Languages

Get StartedView Docs Hugging Face

Claire

Female · English · American

Calm and articulate, with crisp enunciation and a polished finish.

Nora

Female · English · American

Full-bodied and soft, with a calm, relaxed delivery.

Diana

Female · English · American

Measured and professional, with a mature, reassuring presence.

Eleanor

Female · English · British

Refined British pronunciation with an elegant, composed delivery.

Graham

Male · English · American

Mature warmth with a lightly textured tone and quiet assurance.

Leon

Male · English · American

Light and expressive, with quick pacing and flowing phrasing.

Ravi

Male · English · Indian

Smooth mid-range tone with precise Indian English pronunciation.

Evan

Male · English · American

Articulate and easygoing, with clean pronunciation and natural phrasing.

Adrian

Male · Spanish · Castilian

Relaxed and conversational, with spontaneous, friendly energy.

Alba

Female · Spanish · Castilian

Smooth and reassuring, with clear, steady delivery.

Choose what matters most

Our Qwen3-TTS endpoint comes in two variants. Use the same API and change only the ID. Note that output quality is identical.

Standard

For production workloads where serving cost matters most.

Latency
200 ms TTFA
Pricing
$5 / 1M

Fast

For realtime agents and interactive applications where every millisecond matters.

Latency
50 ms TTFA
Pricing
$10 / 1M

Try Nari Model APIs

Create an account to use our public speech models.

Get Started

Faster, cheaper, and competitive quality

LatencyElevenLabs vs Nari FastTTFA
PricingElevenLabs vs Nari StandardUSD / 1M characters

Above
field average

in preliminary
independent benchmarks

Quality

Based on preliminary independent measurements of publicly available TTS endpoints from US East. Results may vary by region, connection, and workload. Pricing checked Sep 8, 2026 from Artificial Analysis.

Built for multilingual speech

Speech generation is currently supported in these 10 languages.

10 Supported Languages

  • Chinese
  • English
  • Japanese
  • Korean
  • German
  • French
  • Russian
  • Portuguese
  • Spanish
  • Italian

Text in. Audio out.

Make one HTTPS request, receive a WAV file, and switch between Standard and Fast with a model ID.

View Docs

cURL

curl --fail --silent --show-error \
  https://api.narilabs.com/v1/audio/speech \
  -H "Authorization: Bearer $NARI_API_KEY" \
  -H "Content-Type: application/json" \
  --data '{
    "model": "qwen3-tts-fast",
    "input": "Hello from Nari Labs.",
    "voice": "diana",
    "response_format": "wav"
  }' \
  --output speech.wav

Built for voice-first products

Conversational agents

Stream spoken responses with low time to first audio.

Narration

Generate long-form speech with consistent voices.

Interactive media

Create responsive dialogue and voiceovers at runtime.

Put realtime speech into production

Get started, read the docs, or talk to our engineers.

Get StartedTalk to an Engineer