Render many short clips once and replay them (prompts, notifications, IVR menus)
Four or more companies' docs show a full fit.
What counts as done
All items rendered in one scripted run within the plan's rate limit, each as a file the agent can store and replay at no further charge, with the same voice across the batch and the total cost known before the run.
How agents ask for it: generate these 200 reminder clips as MP3
, make the hold-music announcements
, pre-render the menu prompts
Does the job 8
The docs show it meets the test above, and nothing we read says otherwise.
| Listing | Why |
|---|---|
| Cartesia Sonic TTS API + MCP B Cartesia AI, Inc. | About 1 credit a character, a caching guide for stock phrases, Free with 2 concurrent requests and paid plans queueing up to ten times the limit; no card is mentioned for Free. source |
| Deepgram Text-to-Speech (Aura-2, Flux TTS) A Deepgram, Inc. | POST /v2/speak batch is documented for IVR prompts and notifications as mp3, opus, flac or aac, at $0.045 per 1,000 characters with $200 of credit and no card. source |
| ElevenLabs Text to Speech API + MCP BB Eleven Labs Inc. | Priced per 1,000 characters ($0.08 v4 and v3, $0.04 Turbo, Flash and v3 Conversational) with plan credits, so a run is costed in advance; the pricing page also sets each plan's concurrency. source |
| Fish Audio TTS API B Hanabi AI Inc. | Limits are concurrency slots, not requests a minute (5 below USD 100 prepaid), the price per UTF-8 byte per model is published, s2.1-pro-free costs nothing for a trial, and a saved reference_id holds the voice across the batch. source |
| Inworld Realtime TTS-2 C Theai, Inc. dba Inworld AI | The Batch API runs many requests as one job at 20 per cent below the realtime rate, $25 per million characters on On-Demand with 70 free minutes and no card, 5 concurrent generations, and each result is a file the agent stores. source |
| Murf TTS API + MCP BB Murf Inc. | Stream Speech returns MP3, WAV or ULAW per call at 1 cent per 1,000 characters on Falcon 2.1, so the total is known in advance; concurrency 5 on US-East with usage returned in each response. source |
| Rime TTS API + MCP C Rime Labs, Inc. | HTTP synthesis returns a clip per call, the Free plan gives $50 of usage a month with no card, and Flex is $0.03 or $0.05 a minute at about 1,000 characters a minute, so a batch cost is computable before the run. source |
| VieNeu-TTS v3 Turbo C Phạm Nguyễn Ngọc Bảo, an individual (named on the card as author; Hugging Face account pnnbao-ump, GitHub account pnnbao97). No company found | infer_batch renders many texts in one batched forward on GPU or sequentially on CPU, the server limits by stream slots with 429 and Retry-After, cost is the hardware, and a named preset voice holds across the batch. source |
Does the job, with a limit 15
It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.
| Listing | Why |
|---|---|
| Amazon Polly BB Amazon Web Services, Inc. | 80 requests a second on standard voices, cached replays free and $4 per million characters make a 2,000-clip run cheap and quick (acc-34 does this), but AWS may ask new accounts for a payment method and the monthly free characters apply only to accounts opened before 15 July 2025. source |
| Audio8-TTS-Realistic 0.6B D Edge0 (edge0.ai), which names no legal entity or location on its site. The technical report's authors give their affiliation as AutoArk, and commit metadata in Edge0-AI/Audio8_TTS uses an autoark.ai domain | The processor takes lists of texts and the README links a batch inference guide at no per-item cost, but the context notes Transformers must stay below 5 to avoid all-zero codes, and no rate figures exist for the adapter. source |
| Azure AI Speech text-to-speech BB Microsoft Corporation | 30 transactions a second and per-character meters make the run and its cost predictable, but an Azure subscription needs a card even for F0, and batch synthesis is not available on F0. source |
| Fish Audio Drama 3 C 39 AI, Inc. (the licensor named in the Fish Audio Community License). fish.audio's terms of use name Hanabi AI Inc. as the provider of the hosted services | Rendering locally costs nothing per item and the API server takes one text at a time with no rate cap, but acc-31's non-deterministic failures mean every clip needs checking, and the 48 GB NVIDIA requirement is the only cost. source |
| Fish Audio S2 Pro D 39 AI, Inc. (the licensor named in the Fish Audio Research License). fish.audio's terms of use name Hanabi AI Inc. as the provider of the hosted services | Local rendering has no rate cap or per-item cost and a saved reference_id holds the voice, but IVR and notification prompts for a business fall outside the research and non-commercial licence. source |
| Google Cloud Text-to-Speech B Google LLC | Standard voices are $4 per million after 4 million free and Chirp 3 HD $30 after 1 million, with 200 requests a minute on Chirp 3, but billing and a card must be enabled before the free allowance can be used. source |
| HeyGen Voice BB HeyGen Technology, Inc. | POST /v3/models/audio/tts returns one WAV or MP3 per call that the agent stores, and Turbo Free costs nothing until 30 November 2026, but 30 requests a minute per member makes a 2,000-item batch over an hour and the paid rate is not on the pages read. source |
| Ming-3.0-tts-0.5B C inclusionAI (the LICENSE's copyright holder; source headers read Copyright (c) Ant Group) | A Python loop over infer.py renders clips free of charge in a fixed built-in voice with deterministic defaults, but the publisher's tested hardware is H800 and H20, so an agent on a small GPU has no stated footing. source |
| Ming-3.0-tts-tiny C inclusionAI, Ant Group (named in the site footer of inclusion-ai.org; the LICENSE copyright line reads inclusionAI) | Free, deterministic with seed and temperature 0, and the served endpoint takes one request per clip, but the stated hardware is an H800, H100 or H200. source |
| MiniMax Speech (T2A) C Nanonoble Pte. Ltd. | Each sync call returns audio the agent stores and the price is published per million characters, but the 60-requests-a-minute cap turns a 2,000-item batch into a 33-minute job and no free allowance exists to test a voice first; acc-40 ships it behind a user's own key. source |
| OmniVoice C k2-fsa, the Next-gen Kaldi open-source project. The paper's authors are at Xiaomi Corp.; no company is named as the licensor | omnivoice-infer-batch renders a JSONL of items across several GPUs at no per-item cost with one saved voice, but IVR or product prompts are commercial use the CC-BY-NC weights do not grant. source |
| Qwen3-TTS 12Hz 1.7B CustomVoice C Alibaba Cloud (the copyright holder in the Qwen licences; RDAP names Alibaba Cloud (Singapore) Private Limited as registrant of qwen.ai, from the first batch's check) | Lists of texts, languages, speakers and instructions render in one call at no charge with the same preset voice, but no GPU memory or throughput is stated and acc-19 reports slow generation. source |
| Resemble AI Text-to-Speech API D Resemble AI, Inc. | The synchronous endpoint returns one base64 clip per call at 40 requests a second per token, which suits a scripted run, but www.resemble.ai forbids this crawler so the per-second price could not be read and the total cannot be known in advance. source |
| Smallest.ai Lightning TTS B Smallest, Inc. | Sync endpoint at $0.175 or $0.195 per 10,000 characters, 250 characters a request and 1 concurrent request, so a batch runs serially; lightning_v3.1_pro changes checkpoint in place on 28 November 2026 and the dated preview id lasts 30 days. source |
| Soniox Text-to-Speech BB Soniox Inc. | REST at 100 requests a minute and 3 concurrent, voice stable by name, but the price is per output audio token so the total is an estimate (about $0.70 an hour), no free credit, and tts-rt-v2 is the only model with no dated snapshot. source |
What stops agents finishing
- a card required before any call, even on the free tier (Azure F0, AWS after 15 July 2025)
- a request-per-minute cap that makes a 2,000-item batch an hour-long job (MiniMax 60 a minute, HeyGen 30 a minute)
- no free allowance to test the voice first (MiniMax)
Around it
Before: database.query-rows. After: storage.upload-file, voice-agents.play-prompt.
Where this comes from
Weight 5 of 100 in Text-to-speech: 4 of 46 accounts tagged (acc-34, 40, 44, 46); the oldest use of these APIs and one Polly's docs lead with, but few people write about it
The same as data: /api/v1/use-cases/text-to-speech.batch-short-clips.json, or the MCP tool get_use_case with id text-to-speech.batch-short-clips. Other jobs in this category: /api/v1/niches?category=text-to-speech.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.