If you are making spoken content, the answer is ElevenLabs for voice (Starter $6/mo), Suno for music (Pro $8/mo) and Descript for editing (Hobbyist $16/mo on annual billing) — about $30/month for the complete creator toolkit. If you are taking notes in meetings, it is Otter.ai for general use ($8.33/mo annual), Granola when a visible recording bot is awkward ($14/user/mo) and Fireflies.ai for sales teams who need CRM sync. If you record anywhere noisy, add Krisp (Core, $8/user/mo annual). If you are buying a voice API rather than an app, skip to the rate table — two prices change this quarter.
Our Pick: ElevenLabs
The voice AI that solved synthetic speech. Audiobooks, podcasts, video voice-overs — if you’ve heard AI-generated voice in the last year, it was probably ElevenLabs. Starter at $6/month unlocks commercial licensing and instant voice cloning. Professional voice cloning (PVC) on Creator tier produces clones nearly indistinguishable from the source.
It also holds the quality crown on measured output, not just on product. Google took it for five days in September; on 28 September 2026 ElevenLabs shipped Eleven v4 and v4 Turbo with 90+ languages (up from 70), cloning from about 10 seconds of audio, and no. 1 on Artificial Analysis’ Speech Arena at 1319 Elo — ahead of Cartesia Sonic 3.6 (1276) and Gemini 3.8 Flash TTS (1267), reversing Eleven v3’s rank-17 position. ElevenLabs reports blind-preference win rates of 65-81% against competing models.
The regional point that decides it for a lot of readers: ElevenLabs clones voices in the EEA and UK behind a spoken consent passphrase, while Gemini’s voice replication is unavailable in the EEA, the UK, Switzerland, India, Illinois and Texas. If you are in Europe and you need a cloned voice, the cheaper option is not an option.
- Price: Free, Starter $6/mo, Creator $22/mo, Pro $99/mo, Scale $299/mo, Business $990/mo
- Ease of Use: 9/10 · Quality: 10/10 · Value: 8/10
- Best for: Podcast production, audiobook narration, video voice-overs, multi-language dubbing, voice cloning
- Not ideal for: Music generation (use Suno), interactive conversational voice (ChatGPT voice), live phone agents (see below)
Building a phone or support agent is a different purchase. OpenAI’s GPT-Live-1 at a flat $0.05 per minute listens and speaks in one model with turn-taking down to 0.798s, and bills wall-clock seconds — so unlike the old audio-token meters, a hold queue costs the same as conversation. Price it before building on ElevenLabs’ conversational stack; OpenAI’s legacy audio and realtime models retire 20 January 2027.
Runner Up: Suno
The AI music leader with 2M+ paid subscribers. Suno v5 is the quality benchmark — ELO 1,293 puts it ahead of every competitor. Full songs with vocals, structure, and emotional arc. The free tier gives 50 credits/day but no song downloads, which is the limit that matters: you can generate but not take anything away. Pro at $8/month is the standard paid tier, with 2,500 credits, 20 downloads and commercial licensing.
- Price: Free (50 credits/day, no downloads), Pro $8/mo, Premier $24/mo
- Ease of Use: 9/10 · Quality: 10/10 · Value: 8/10
- Best for: Content creator background music, filmmakers needing soundtracks, songwriters experimenting
- Not ideal for: Audio mixing/mastering (not its purpose), voice-over work (use ElevenLabs)
The case for Lyria instead. Google brought Lyria 3.5 to the Gemini API on 4 September at $0.08 per song — about 5x Suno Pro’s effective rate after Suno’s cut, so it does not displace Suno for making music by hand. It wins two cases: Suno still sells no public API, so Lyria is the only frontier music model you can build a product on under a contract; and every Lyria output carries a SynthID mark at the model, while Suno’s own watermarking, announced 6 August 2026, was still described as rolling out. If you publish commercially into the EU, that beats the price gap.
Budget Pick: Descript
Edit spoken content by editing text. Import a recording, Descript transcribes it, and editing the transcript edits the audio. Remove “um”s with one click. Studio Sound cleans audio. Free tier gives 1 hour of media/month and 100 one-time AI credits.
The monthly/annual gap here is unusually wide — up to 35%. Hobbyist at $16/month annual (10 media hours, 400 AI credits) covers a weekly podcast; step up to Creator at $24 annual (30 hours, 800 credits) only when 10 hours stops being enough. Paying monthly makes those $24 and $35.
- Price: Free (1 media hour/mo), Hobbyist $16/mo annual ($24 monthly), Creator $24/mo annual ($35 monthly), Business $50/mo annual ($65 monthly)
- Ease of Use: 8/10 · Quality: 9/10 · Value: 8/10
- Best for: Podcast editing, long-form video, interviews, removing filler words
- Not ideal for: Pure music production, cinematic video work (Premiere better), anyone who would pay monthly rather than annually
For Meeting Notes (mainstream pick): Otter.ai
The polished consumer-friendly meeting AI. 300 free transcription minutes per month, Pro at $8.33/month annual. 93-95% transcription accuracy in clean audio. Otter Chat lets you ask questions across all past meetings — “what did the prospect say about pricing on Tuesday?” — genuinely useful and underadvertised. The friendliest meeting AI to recommend to a non-technical user. Supports Zoom, Google Meet, Teams.
- Price: Free (300 min/mo), Pro $8.33/user/mo annual ($16.99 monthly), Business $19.99/user/mo annual ($30 monthly)
- Ease of Use: 9/10 · Quality: 9/10 · Value: 9/10
- Best for: Solo professionals, knowledge workers, anyone wanting search across meeting history, mainstream small-team workflows
- Not ideal for: Bot-free recording (try Granola), complex sales-team CRM workflows (try Fireflies)
For Privacy-First Meetings: Granola
The bot-free meeting AI for people who think during meetings. No visible recording bot joins your call — Granola listens locally on macOS/Windows/iOS while you type sparse notes; AI fleshes them into structured summaries with verbatim transcript linking. $1.5B valuation, $125M Series C in March 2026. Best for executives, consultants, lawyers, and anyone where the visible-bot moment is awkward or compliance-problematic.
The free tier is now limited by retention rather than meeting count — 30 days of history, replacing the old lifetime cap of 25 meetings. Better for occasional use, worse as a permanent archive, which is the usual reason to pay.
- Price: Free (30-day meeting history), Business $14/user/mo, Enterprise $35/user/mo
- Ease of Use: 9/10 · Quality: 9/10 · Value: 8/10
- Best for: Executives in 6+ meetings/day, consultants in client calls, regulated industries, “I think during meetings” power users
- Not ideal for: Sales/CS teams needing CRM integrations (Fireflies better), Chromebook/Android-first users (no support)
For Sales / Customer Success Teams: Fireflies.ai
The meeting AI built for revenue-team workflows. AskFred for conversational queries across meeting history, Talk to Fireflies for in-call AI assistance, 100+ language transcription, deep CRM integrations (Salesforce, HubSpot, Pipedrive). The reason to pick it over Otter is the CRM write-back, not the note quality.
Fireflies stopped publishing its prices. As of 7 October 2026 its pricing page names Free, Pro, Business and Enterprise and says paid plans bill per seat, but shows no figures — treat the rates below as last-known and get a quote. That opacity is itself a reason to prefer Otter or Granola if you are buying without a procurement process.
- Price (last published): Free, Pro $10/user/mo annual, Business $19/user/mo, Enterprise custom
- Ease of Use: 8/10 · Quality: 9/10 · Value: 9/10
- Best for: Sales teams, customer success, CRM-integrated workflows, multi-language transcription
- Not ideal for: Solo professionals (Otter simpler), bot-free contexts (Granola)
Read full Fireflies.ai review →
For Background Noise + Meeting Quality: Krisp
The AI noise cancellation that became the meeting-quality default. Removes background noise, echo, and voice tone artifacts in real time across Zoom, Teams, Google Meet, and any other meeting tool. Adds AI meeting notes and transcription on the paid tiers. For anyone working from a noisy environment or recording from non-studio setups, Krisp is the upstream quality fix that everything else benefits from.
The tiers are now Core and Advanced (previously Pro and Business), the annual discount is the steepest here at 50%, and entry is a 7-day full-feature trial rather than a perpetual free tier — Krisp retired the old 60-minutes-a-day free plan. Core and Advanced differ mainly on accent conversion: 1 hour daily against 4.
- Price: 7-day free trial, Core $8/user/mo annual ($16 monthly), Advanced $15/user/mo annual ($30 monthly)
- Ease of Use: 10/10 · Quality: 9/10 · Value: 9/10
- Best for: Remote workers in noisy environments, podcasters recording at home, anyone who joins meetings from cafes/airports
- Not ideal for: Users who don’t have meeting noise problems
Quick Comparison
| Tool | Entry paid price | Free tier | Ease | Quality | Value | Best For |
|---|---|---|---|---|---|---|
| ElevenLabs | $6/mo Starter | 10,000 credits/mo | 9/10 | 10/10 | 8/10 | AI voice generation |
| Suno | $8/mo Pro | 50 credits/day, no downloads | 9/10 | 10/10 | 8/10 | AI music generation |
| Descript | $16/mo annual Hobbyist | 1 media hour/mo | 8/10 | 9/10 | 8/10 | Podcast/video editing |
| Otter.ai | $8.33/user/mo annual Pro | 300 min/mo | 9/10 | 9/10 | 9/10 | Mainstream meeting notes |
| Granola | $14/user/mo Business | 30-day history | 9/10 | 9/10 | 8/10 | Bot-free meetings |
| Fireflies.ai | Not published | Yes, unspecified | 8/10 | 9/10 | 9/10 | Sales/CS teams |
| Krisp | $8/user/mo annual Core | 7-day trial only | 10/10 | 9/10 | 9/10 | Noise cancellation |
Subscription prices verified 7 October 2026 against elevenlabs.io/pricing, suno.com/pricing, descript.com/pricing, otter.ai/pricing, granola.ai/pricing and krisp.ai/pricing. Fireflies.ai published no figures on that date.
Buying the API, not the app
Vendor pages will not let you compare voice APIs, because no two of them meter the same thing. ElevenLabs and Microsoft bill per character, Google bills audio output tokens, OpenAI’s GPT-Live-1 bills wall-clock seconds, and Google’s Live Avatar bills the video track at nothing at all. The per-million-character figures for Gemini TTS circulating in roundups are estimates, and published conversions disagree by more than 2x.
So here is every rate converted to one basis: one hour of generated speech, taken as ~54,000 characters (about 150 words per minute). Use it to rank, then check your own speech density before committing.
| Model | List rate | Per hour of audio | Provenance mark | Note |
|---|---|---|---|---|
| Qwen-Audio-3.1-TTS | — | ~$0.18 | No | China-hosted Beijing node; not listed on Alibaba’s international rate card |
| Gemini 3.8 Flash-Lite TTS | $6.00/M audio output tokens | ~$0.54 | SynthID | Doubles to ~$1.08 on 1 Jan 2027; no voice cloning in EEA/UK/CH/IN/IL/TX |
| Gemini 3.8 Flash TTS | $9.00/M audio output tokens | ~$0.81 | SynthID | Doubles to ~$1.62 on 1 Jan 2027; same regional cloning limits |
| MAI-Voice-2.1-Flash | $15/M characters | ~$0.81 | Not addressed | Public preview, no SLA; no published expiry on the rate |
| MAI-Voice-2.1 | $22/M characters | ~$1.19 | Not addressed | 150ms end-to-end, 23 languages, 26 locales |
| Eleven v4 Turbo | $40/M characters | ~$2.16 | Yes | $11/M (~$0.59/hr) until 12 Oct 2026 on the 72% launch discount |
| Eleven v4 | $80/M characters | ~$4.32 | Yes | $22/M (~$1.19/hr) until 12 Oct 2026; no. 1 on Speech Arena |
Two things fall out of that table that no vendor page will tell you:
- The cheapest Western rate today is an ElevenLabs discount, and it dies on 12 October 2026. Eleven v4 Turbo at $11/M undercuts MAI-Voice-2.1-Flash’s $15/M right now; next week it is 2.7x more expensive. If you are benchmarking this fortnight, you are benchmarking a price that will not exist in November.
- Microsoft wins the cheap slot on 1 January 2027 by standing still. MAI-Voice-2.1-Flash is level with Gemini 3.8 Flash TTS today and above Flash-Lite — it was widely reported as the cheapest credible non-Chinese voice model, and it is not. But its rates are the only ones here with no published expiry, so when Google’s double it takes the cheapest Western slot, at the cost of a public preview with no SLA and a launch post that addresses neither provenance marking nor regional cloning.
If you are on an older Gemini TTS endpoint, note that gemini-3.1-flash-tts-preview and gemini-2.5-flash-preview-tts both shut down on 17 November 2026, with gemini-3.8-flash-tts as the replacement — so the January price doubling catches you either way. Full dates and a decision table: MAI-Voice 2.1 vs ElevenLabs and the model retirement calendar.
What about transcription prices?
Upstream speech-to-text has fallen ~20x in a year: Gemini 3.5 Transcribe met the $0.30/audio-hour floor in August, Meta’s Muse Voice Transcribe broke it at $0.18 on 1 September, and Qwen-Audio-3.1-ASR reached **$0.016** on a China-hosted node. On 1 October the direction reversed for the first time this year — MAI-Transcribe-2-Streaming took the accuracy crown at 2.5% WER and charges $0.54, three times Meta’s rate for about six fewer wrong words per thousand, on an introductory rate ending 31 December 2026.
Default to Meta unless a single error is expensive — medical, legal, captioning, live agents. For the tools on this page it matters less than you’d think: Otter, Granola and Fireflies compete on calendar integration, history and CRM sync, not raw accuracy, so cheaper transcription reaches you as better transcripts, not a reason to switch.
Which audio tool should you pick?
- For voice-over work: ElevenLabs Creator ($22/mo).
- For custom music: Suno Pro ($8/mo).
- For podcast editing: Descript Hobbyist ($16/mo annual), stepping up to Creator ($24/mo annual) past 10 media hours.
- For meeting notes (general use): Otter.ai Free, then Pro at $8.33/user/month annual.
- For privacy-sensitive client meetings: Granola Business ($14/user/mo).
- For sales / customer success teams: Fireflies — but get a quote, since it no longer lists prices.
- For background noise problems: Krisp Core ($8/user/mo annual).
- For a voice API in the EU: ElevenLabs — Google will not clone a voice for you there at any price.
Most content creators run ElevenLabs + Descript + Suno — the complete audio creation toolkit for podcast, video, and music work. ElevenLabs Starter ($6) plus Suno Pro ($8) plus Descript Hobbyist on annual billing ($16) is $30/month. Pay Descript monthly instead and the same three tools cost $38, so the single highest-leverage billing decision in this guide is committing to Descript for a year. Add Otter.ai or Granola depending on the meeting context.
What changed
- 7 Oct 2026: Restructured answer-first and added the $/hour voice API rate table. Every subscription price re-verified: ElevenLabs Business corrected $1,320 → $990, Suno Pro $10 → $8 and Premier $30 → $24, Descript’s tiers renamed with the $16 slot now Hobbyist on annual billing (Creator is $24 annual / $35 monthly), Krisp’s Pro/Business renamed Core/Advanced with $8 and $15 confirmed as annual rates, Granola’s free tier moved from a 25-meeting lifetime cap to 30-day retention, and Fireflies stopped publishing prices.
- 3 Oct 2026: Microsoft entered voice generation — MAI-Voice-2.1 at $22/M characters and Flash at $15/M, public preview with no SLA — see analysis.
- 1 Oct 2026: MAI-Transcribe-2-Streaming took the ASR accuracy crown from Meta at 2.5% WER and $0.54/audio-hour, an introductory rate ending 31 December 2026.
- 28 Sep 2026: Eleven v4 and v4 Turbo shipped with 90+ languages and retook no. 1 on Artificial Analysis’ Speech Arena at 1319 Elo.
- 25 Sep 2026: Gemini 3.8 Live with Live Avatar went GA with real-time lip-synced video across 97 languages — and Google published no rate for the video track at all.
- 23 Sep 2026: Gemini 3.8 Flash TTS and Flash-Lite TTS shipped at ~$0.81 and ~$0.54 per hour, with voice replication blocked in the EEA, UK, Switzerland, India, Illinois and Texas.
- 22 Sep 2026: Qwen-Audio-3.1-ASR cut streaming transcription ~95% to ~$0.016/audio-hour and changed the meter to tokens.
- 10 Sep 2026: GPT-Live-1 moved voice billing to wall-clock seconds at a flat $0.05/minute; OpenAI’s legacy audio and realtime models retire 20 January 2027.
Frequently asked questions
What is the best AI voice generator in 2026?
ElevenLabs, and as of 28 September 2026 it leads on measured output quality again as well as on product. Eleven v4 and v4 Turbo shipped that day with 90+ languages, and Eleven v4 took the no. 1 slot on Artificial Analysis' Speech Arena at 1319 Elo — ahead of Cartesia Sonic 3.6 (1276) and Gemini 3.8 Flash TTS (1267), and a decisive reversal from Eleven v3's rank-17 position earlier in September. On price it is still the expensive option once its launch discount lapses on 12 October 2026: the API list rate is $80 per million characters for v4 and $40 for v4 Turbo. Gemini 3.8 Flash-Lite TTS is the cheapest provenance-marked rate at about $0.54 per hour of generated audio until Google's rates double on 1 January 2027. Full breakdown: [MAI-Voice 2.1 vs ElevenLabs](/compare/mai-voice-vs-elevenlabs/).
What's the best AI music generator?
Suno. It generates complete songs — vocals, instruments, structure — from text prompts and leads the category in 2026 for both quality and ease of use. Google's Lyria 3.5 is the alternative worth knowing about, because Suno sells no public API and Lyria does.
What's the best AI meeting transcription tool?
Otter.ai is the mainstream pick for meeting notes and transcription. Granola is the privacy-first alternative (notes stay local), and Fireflies.ai fits sales and customer-success teams needing CRM integration. All three compete on calendar integration and meeting history rather than raw accuracy, so the collapsing price of upstream speech-to-text reaches you as gradually better transcripts, not as a reason to switch.
Can AI edit podcasts and audio?
Yes — Descript is the budget-friendly pick. It turns audio editing into text editing: delete a sentence from the transcript and it's cut from the audio, plus studio sound enhancement and filler-word removal.
Is it cheaper to use a voice API than an ElevenLabs subscription?
It depends entirely on volume, and the comparison is hard because vendors meter differently — ElevenLabs and Microsoft bill per character, Google bills audio output tokens, and OpenAI's GPT-Live-1 bills wall-clock seconds. Priced on a common basis of one hour of generated speech, Gemini 3.8 Flash-Lite TTS is the cheapest Western rate at about $0.54/hour, against roughly $4.32/hour for Eleven v4 at list. But two dated cliffs change that: ElevenLabs' launch discount lapses 12 October 2026 and Google's TTS rates double on 1 January 2027. See the rate table below.