1 · Top Score
Eleven v3
ElevenLabs
94.2
Latest general-availability ElevenLabs speech model focused on expressive text-to-speech, multi-speaker control and production-grade voice output.
ttsRealtimeVoice CloneAPI
SWEN Audio Registry
Best AI Audio Models 2026Speech, Voice and Music
This page now uses a canonical audio registry instead of static lists. The ranking covers TTS, STT, voice cloning and music generation with a modality-specific score.
6
canonical products ranked
5
with public API or developer access
2
with voice cloning or programmable voice
2 · Top Score
GPT-4o Transcribe
OpenAI
OpenAI91.4
OpenAI speech-to-text model optimized for more accurate transcription than older Whisper-based production defaults.
sttAPI
3 · Top Score
Speech 2.8
MiniMax
90.6
MiniMax speech stack with HD and turbo variants, native sound-tag support and high-fidelity cloning for developer-facing audio generation.
ttsRealtimeVoice CloneAPI
Full Ranking
Composite score built from quality, latency, control, value and API readiness.
0 open source
| # | Model | Company | Score | Quality | Latency | Control | Price | Highlights | Release |
|---|---|---|---|---|---|---|---|---|---|
| 1 | Eleven v3 Latest general-availability ElevenLabs speech model focused on expressive text-to-speech, multi-speaker control and production-grade voice output. | ElevenLabs | 94.2 | 97 | 91 | 96 | Paid plans from US$5/month | Primary use: Expressive TTS • Realtime: Yes • Languages: 70+ | Feb 2026 |
| 2 | GPT-4o Transcribe OpenAI speech-to-text model optimized for more accurate transcription than older Whisper-based production defaults. |  OpenAI | 91.4 | 95 | 84 | 82 | Pay-as-you-go via Transcription API | Primary use: High-accuracy STT • Context: 16k • Realtime: No | — |
| 3 | Speech 2.8 MiniMax speech stack with HD and turbo variants, native sound-tag support and high-fidelity cloning for developer-facing audio generation. | MiniMax | 90.6 | 93 | 90 | 92 | US$60/M chars (turbo) or US$100/M chars (HD) | Primary use: Realtime + HD TTS • Voice cloning: Yes • Sound tags: Native | — |
| 4 | GPT-Realtime-Whisper Streaming speech-to-text model from OpenAI for low-latency transcript deltas and live audio applications. | OpenAI | 89.7 | 90 | 96 | 80 | Priced by audio duration | Primary use: Realtime STT • Context: 16k • Realtime: Native | — |
| 5 | Suno v4.5 Latest major Suno music-generation model tier focused on richer vocals, more accurate style following and consumer-grade music creation speed. | Suno | 88.9 | 92 | 84 | 87 | Paid plans from creator tiers | Primary use: Music generation • Commercial use: Plan-dependent • API: No public API | May 2025 |
| 6 | Music 2.6 MiniMax music generation model in the same flagship family as Speech 2.8, aimed at API-driven music workflows rather than UI-only creation. | MiniMax | 87.3 | 89 | 82 | 88 | API-accessible via MiniMax platform | Primary use: Programmable music gen • API: Yes • Suite: Speech & Music | — |
Score Breakdown
Qualitytop model reference
Latencytop model reference
Controltop model reference
FAQ
Why is this page no longer using editorial fixed lists?
Because the audio page now reads from the canonical SWEN Audio Registry. New speech, transcription and music entries can be refreshed automatically by the benchmark sync stack instead of requiring manual page edits.
What does the composite score optimize for?
The score balances output quality, latency, controllability, value and API readiness. That prevents a consumer-only product from outranking a production-ready API stack on hype alone.
Can music generators appear together with TTS and STT models?
Yes. This page is modality-first, not task-first. Audio includes speech synthesis, speech recognition and music generation, while each model keeps its own subcategory and descriptor set.