TheRouter.aiTheRouter.ai
Back to Models

TTS-1 HD

openaiopenai/tts-1-hd

Text-to-speech model optimized for quality.

TTS-1 HD model reference ↗Text-to-speech API guide ↗Audio API reference (v1/audio/speech) ↗DevDay announcement (November 6, 2023) ↗OpenAI pricing ↗

TTS-1 HD is OpenAI's high-fidelity text-to-speech model, launched alongside TTS-1 at OpenAI DevDay on November 6, 2023. It converts text into natural-sounding speech optimised for audio quality rather than latency. At $30.00 per million characters — double the price of TTS-1 ($15.00/M chars) — it is the preferred option for content where audio fidelity is the primary concern: audiobooks, podcasts, e-learning narration, and broadcast media. It shares the same voice set (alloy, ash, coral, echo, fable, onyx, nova, shimmer) and output formats (MP3, Opus, AAC, FLAC, WAV, PCM) as TTS-1, differing only in the quality/latency trade-off.

As of 2025–2026, OpenAI recommends gpt-4o-mini-tts for new voice-agent projects because it adds instruction-following (controlling tone, accent, and speaking style via a free-text instructions parameter). TTS-1 HD remains fully supported and is the right choice for production pipelines that generate pre-rendered audio assets — scenarios where the request is not on the interactive critical path and an extra few hundred milliseconds per synthesis is acceptable in exchange for noticeably crisper output. It is callable via TheRouter's OpenAI-compatible endpoint (/v1/audio/speech) with no client changes required.

Best for
  • • Audiobooks and long-form narration where listeners will judge every sentence — TTS-1 HD's higher fidelity is perceptible over extended listening sessions
  • • Podcast production and marketing videos where audio will be published or distributed — the quality difference over TTS-1 is audible to a non-technical audience
  • • E-learning and corporate training narration where content will be played back repeatedly — reduced audio artefacts improve perceived professionalism
  • • Batch pre-rendering of audio assets (product demos, explainer videos, IVR recordings) where synthesis happens offline and cost-per-request is predictable
Reach for something else if
  • • Real-time conversational agents where first-audio latency is critical — use TTS-1 for lower latency, or gpt-4o-realtime-preview for native streaming voice
  • • New voice-agent builds requiring prompt-controlled speech style (tone, accent, emotion) — use gpt-4o-mini-tts instead, which supports the instructions parameter
  • • Cost-sensitive high-volume pipelines where the quality premium cannot be justified — TTS-1 at $15/M chars delivers acceptable quality at half the cost
Context Length
--
Max Output
--
Text Priceper 1M characters
$32.40/ 1M chars

Modalities

text→audio

Capabilities

TTS

Media Generation Capabilities

tts
output_formats
  • mp3
  • opus
  • aac
  • flac
voices
  • alloy
  • echo
  • fable
  • onyx
  • nova
  • shimmer
defaults
voice
alloy

Pricing Breakdown

TypeRate
Text$32.40 / 1M chars

request is price per 1M characters

Supported Parameters

inputvoiceresponse_formatspeed

Specifications

Release date2023-11-06 (OpenAI DevDay)openai.com ↗verified
Pricing$30.00 per 1 M charactersopenai.com/api/pricing ↗verified
Max input length4 096 characters per requestplatform.openai.com ↗verified
Supported voicesalloy, ash, coral, echo, fable, onyx, nova, shimmer (9 voices)platform.openai.com ↗verified
Output formatsMP3 (default), Opus, AAC, FLAC, WAV, PCMplatform.openai.com ↗verified
Speed control0.25–4.0× (default 1.0×)platform.openai.com ↗verified
Latency vs TTS-1Higher latency than TTS-1 — optimised for quality over speedplatform.openai.com ↗verified
Instruction-following (voice style)Not supported — use gpt-4o-mini-tts for prompt-controlled speechplatform.openai.com ↗verified
Rate limits (Tier 1)500 RPMplatform.openai.com ↗verified
Training cutoffNot publicly disclosedunknown
LicenseProprietary (OpenAI API Terms of Service)openai.com/policies ↗verified

Benchmarks

BenchmarkDistributionScoreSource
Relative audio quality vs TTS-1
OpenAI describes TTS-1 HD as delivering higher audio quality than TTS-1 at the cost of higher latency. No numeric MOS or CMOS figures are publicly disclosed.
Higher quality than TTS-1qualitativeplatform.openai.com ↗

API Usage Examples

Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.

cURL
curl https://api.therouter.ai/v1/audio/speech   -H "Content-Type: application/json"   -H "Authorization: Bearer $THE_ROUTER_API_KEY"   -d '{
    "model": "openai/tts-1-hd",
    "input": "Welcome to TheRouter.",
    "voice": "alloy"
  }'

API guide

Full API reference →

Speech generation (v1/audio/speech)

cURL
curl https://api.therouter.ai/v1/audio/speech \
  -H "Authorization: Bearer $THEROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/tts-1-hd",
    "input": "The quick brown fox jumped over the lazy dog.",
    "voice": "nova"
  }' \
  --output speech-hd.mp3

More from openai

openai/tts-1openai/gpt-4o-mini-ttsopenai/gpt-4o-realtime-previewopenai/gpt-audio

Similar models

Cross-provider sibling models
mistral/voxtral-tts
fishaudio/fish-speech-1.5
indexteam/indextts-2
alibaba/cosyvoice2-0.5b

News & changes

2025-03-20

OpenAI launches gpt-4o-mini-tts with instruction-following for voice agents

OpenAI released gpt-4o-mini-tts alongside new speech-to-text models, positioning it as the recommended model for new voice agent projects. Unlike TTS-1 HD, it accepts a free-text instructions field for controlling accent, tone, emotion, and speaking style. TTS-1 HD remains supported and is still the preferred choice for quality-first pre-rendered audio where instruction-following is not needed.

re-authored by TheRouteropenai.com ↗
2024-03-30

OpenAI TTS voices arrive on Azure OpenAI Service

Microsoft announced both TTS-1 and TTS-1 HD on Azure OpenAI Service and Azure AI Speech, extending enterprise-grade, data-residency-compliant access to OpenAI's highest-fidelity TTS offering. This makes TTS-1 HD available to regulated industries — healthcare, finance, government — that require data to remain within Azure's geographic boundaries.

re-authored by TheRoutertechcommunity.microsoft.com ↗
2023-11-06

OpenAI DevDay: TTS-1 and TTS-1 HD launched

OpenAI introduced its first production text-to-speech models at DevDay 2023: TTS-1 for low-latency streaming and TTS-1 HD for high-fidelity output. Both models support 6 initial voices (alloy, echo, fable, onyx, nova, shimmer), multiple output formats, and the OpenAI Audio API (/v1/audio/speech), enabling any application to add natural-sounding voice generation with a single API call.

re-authored by TheRouteropenai.com ↗

Frequently asked

What is the difference between TTS-1 HD and TTS-1?

TTS-1 HD is optimised for audio quality and is better suited for content that will be listened to carefully — podcasts, audiobooks, corporate narration. TTS-1 is optimised for low latency and is better for real-time or streaming applications. TTS-1 HD costs $30/M characters; TTS-1 costs $15/M characters. Both support the same voices, output formats, and speed control.

re-authored by TheRouterplatform.openai.com ↗
Should I use TTS-1 HD or gpt-4o-mini-tts for new projects?

Use gpt-4o-mini-tts for any new voice-agent or conversational build — it adds instruction-following (tone, accent, personality via instructions) and is OpenAI's recommended path. Use TTS-1 HD when you need the highest audio quality for pre-rendered, non-interactive content and do not need style control — audiobooks, podcast segments, or corporate e-learning where the synthesis runs offline.

re-authored by TheRouterplatform.openai.com ↗
How do I use TTS-1 HD via TheRouter?

Point your client's baseURL to https://api.therouter.ai/v1, use your TheRouter API key, and set model to openai/tts-1-hd. All other parameters (voice, speed, response_format, input) remain identical to the OpenAI SDK. TheRouter handles routing, billing, and rate-limit aggregation transparently.

re-authored by TheRouterplatform.openai.com ↗
Which output format should I use with TTS-1 HD?

MP3 is the default and works well for most web and mobile delivery. For content that will be post-processed in a DAW or edited further, use FLAC (lossless) or WAV. Opus offers smaller file sizes than MP3 at equivalent quality, making it good for bandwidth-constrained delivery. PCM is best for direct audio pipeline integration (e.g., telephony or real-time mixing).

re-authored by TheRouterplatform.openai.com ↗
Fact ledger — every claim on this page traces here
sourceURLretrieved
Release dateopenai.com ↗2026-06-09verified
Pricingopenai.com/api/pricing ↗2026-06-09verified
Max input lengthplatform.openai.com ↗2026-06-09verified
Supported voicesplatform.openai.com ↗2026-06-09verified
Output formatsplatform.openai.com ↗2026-06-09verified
Speed controlplatform.openai.com ↗2026-06-09verified
Latency vs TTS-1platform.openai.com ↗2026-06-09verified
Instruction-following (voice style)platform.openai.com ↗2026-06-09verified
Rate limits (Tier 1)platform.openai.com ↗2026-06-09verified
Training cutoff——unknown
Licenseopenai.com/policies ↗2026-06-09verified
Relative audio quality vs TTS-1platform.openai.com ↗2026-06-09verified
OpenAI launches gpt-4o-mini-tts with instruction-following for voice agentsopenai.com ↗2026-06-09verified
OpenAI TTS voices arrive on Azure OpenAI Servicetechcommunity.microsoft.com ↗2026-06-09verified
OpenAI DevDay: TTS-1 and TTS-1 HD launchedopenai.com ↗2026-06-09verified
What is the difference between TTS-1 HD and TTS-1?platform.openai.com ↗2026-06-09to verify
Should I use TTS-1 HD or gpt-4o-mini-tts for new projects?platform.openai.com ↗2026-06-09to verify
How do I use TTS-1 HD via TheRouter?platform.openai.com ↗2026-06-09to verify
Which output format should I use with TTS-1 HD?platform.openai.com ↗2026-06-09to verify
Help & contact