Respeecher should be evaluated first as a speech-to-speech and licensed-voice production system, not as a generic text-to-speech catalogue. Its self-serve Marketplace combines TTS, STS, API and a Pro Tools workflow; its separate Voice Lab handles custom, sales-led studio and enterprise projects. Mixing those two purchasing paths produces misleading conclusions about price, rights and support.
> Distinctive strength: Speech-to-speech can retain a consented performance's timing and delivery while transforming it into another licensed voice. > > Where it stops being an advantage: Commodity narration rarely justifies the added performance, consent-contracting and studio-integration complexity.
The strongest reason to shortlist Respeecher is a workflow in which a consented performance must become another licensed voice while retaining timing and delivery. The weakest reason is simply wanting inexpensive commodity narration. In that case, paying for STS, consent contracting and studio integration may add complexity without improving the actual job.
Respeecher's output and support have not been tested hands-on by BenPicks. Current pricing, API, terms and ethics material establish the buying boundaries; “broadcast quality” remains a vendor claim until a blind test with your actors, languages and production chain reproduces it.
Marketplace or Voice Lab: the decision in 60 seconds
| Question | Source-led answer |
|---|
| Self-serve product | Voice Marketplace: TTS, STS, API, projects and eligible Pro Tools use. |
| Custom production | AI Voice Lab: sales-led voice creation and production support. |
| Creator | $89 monthly; 400k TTS characters + 90 STS minutes. |
| Power | $499 monthly; 3M TTS characters + 900 STS minutes. |
| PAYG | $5–$250 packs combining listed TTS characters or STS minutes. |
| Trial boundary | Free evaluation advertised; commercial projects on trial are prohibited. |
| Rights gate | Commercial use depends on offer, chosen voice and project/terms—not possession of an audio file. |
| Main unknowns | Full retention/deletion, technical revocation, enterprise assurance and binding SLA/support. |
Is Marketplace the same as Voice Lab?
No. Marketplace is the accessible account product with published packs and subscriptions. It supports a library of voices, TTS and STS, API keys, projects and conversion exports. Voice Lab is positioned for film, television, games, music and enterprise custom voices, with a sales process and production involvement.
Choose the path before evaluating. A Marketplace trial cannot establish Voice Lab delivery, contract rights or engineering support. Conversely, a custom studio case study does not prove what a self-serve subscription includes. Ask every salesperson and reviewer to label evidence “Marketplace” or “Voice Lab.”
How much does Respeecher Marketplace cost?
Published pay-as-you-go packs are:
| Price | Credits | TTS characters | STS minutes |
|---|
| $5 | 5 | 20,000 | 5 |
| $15 | 16 | 60,000 | 16 |
| $27 | 30 | 120,000 | 30 |
| $70 | 100 | 400,000 | 100 |
| $250 | 500 | 2,000,000 | 500 |
Subscription prices shown at review time:
| Plan | Regular monthly price | Yearly display | Included usage |
|---|
| TTS only | $18 after $9 first month | $14/month equivalent | Character selection varies in the pricing control |
| Creator | $89 after $44.50 first month | $74/month equivalent | 400k TTS + 90 STS minutes |
| Power | $499 after $249.50 first month | $414/month equivalent | 3M TTS + 900 STS minutes |
Do not turn the first-month promotion into the long-term cost. At regular monthly price, Creator is about $0.2225 per 1,000 included TTS characters or $0.989 per included STS minute if one allowance is treated in isolation. Power is about $0.1663 per 1,000 characters or $0.554 per STS minute. These are allocation ratios, not promised marginal billing rates, because each subscription bundles both units.
How do corrections change the cost?
Speech-to-speech bills processed material, not only accepted masters. A 90-minute Creator workload with 25% regenerated needs 112.5 processed minutes; at 50%, 135. That exceeds the included 90 before considering failed conversions or alternative voices.
Track four figures separately: source minutes, processed minutes, regenerated minutes and accepted master minutes. For TTS, track submitted characters by version. Then calculate:
`subscription + packs + labour + editing ÷ accepted finished minutes`
This makes a cheaper but correction-heavy voice comparable with a higher-priced voice that passes earlier. The pricing page alone cannot answer which one wins.
Can trial output be used commercially?
No. Respeecher advertises free evaluation, but Marketplace terms explicitly prohibit commercial projects on trial plans. Use trial output only to decide whether a paid, correctly licensed workflow is worth testing.
Pricing marks commercial-use availability per offer, while the Marketplace agreement still controls access, acceptable use and voice-specific restrictions. Retain the plan receipt, terms version, selected voice and project scope with every master. Do not infer that a generic commercial checkmark permits celebrity impersonation, political use, sublicensing or a different client/project.
How does Respeecher handle voice consent?
Respeecher's ethics and production pages state that explicit permission from the voice owner or estate is required for replicated/custom voices. Current consent guidance recommends naming the technology, project/media, territory, language, term, compensation, prohibited contexts, revocation and a separate training-data clause.
That is a useful baseline, not legal automation. Your signed agreement must identify who can request lines, approve them, receive models/files and publish output. It should explain what happens after withdrawal to the model, source recordings, backups and already distributed content.
Public documentation reviewed here did not establish a complete technical revocation and downstream-output procedure for every Marketplace voice. Treat contract scope and system controls as two separate gates; passing one does not prove the other.
What does the API actually support?
The documented Marketplace flow is production-oriented:
- Authenticate with an expiring API key or session cookie/CSRF flow.
- Create projects and folders.
- Upload WAV, OGG, MP3 or FLAC for STS, or create a TTS recording from text.
- Optionally calibrate mean pitch for transformation.
- Choose voice, narration style, accent and pitch correction where applicable.
- Submit conversion orders, inspect status and retry failures.
- Download WAV conversions or export project/recording ZIP archives, optionally starred-only.
Default per-user rate limits are 500 “fast” and 100 “slow” requests every 300 seconds. Upload, TTS creation, conversion, redo and calibration fall into slow processing. A 429 response includes `Retry-After`, and increased limits require a request.
Design queues around those limits. Do not launch retries without idempotency and a spending/allowance guard, or a transient failure can become duplicate processing.
Does the Pro Tools plugin replace the web workflow?
The documented plugin is beta. It renders speech-to-speech inside Pro Tools and requires a plan with STS; TTS-only is excluded. The page lists TTS and singing conversion in the plugin as “coming soon,” so do not buy for those unshipped plugin functions.
Test it with the exact Pro Tools/OS versions, session format, audio length and collaboration method. Confirm whether renders, calibration and credentials travel safely between editors and whether failures can be reproduced outside the plugin through the API.
What happens to recordings and voice models?
Voice Lab says private client content is not used to train public models, and game-production material describes custom models as project-scoped under agreed consent. Those are meaningful but narrow statements; they should not be stretched into a universal retention or training answer for every Marketplace feature.
The API has delete endpoints for recordings/projects and exposes model/account objects, but the retained public sources did not establish complete timelines for source audio, conversions, custom models, logs and backups. Nor did they provide a full public procurement pack covering SOC/ISO evidence, SSO/SCIM, encryption scope, DPA/subprocessors, incident notification and audit rights.
Request those answers before uploading unreleased performances or regulated/client-confidential material. Test deletion with uniquely identifiable dummy audio, then obtain written confirmation of what the operation covers.
When preserving a consented performance justifies Respeecher
Shortlist it when speech-to-speech identity transformation, actor/estate consent and integration with audio production are central. It can be especially relevant for dialogue pickups, character continuity, localization and high-iteration game/film work where timing and performance matter.
Look elsewhere when you need only low-cost narration, a simple browser editor, a transparent one-unit price or self-serve enterprise security evidence. Also pause if the selected voice's license or revocation terms are unclear, regardless of output quality.
Use the AI Voice directory to separate studio and API products, check the commercial-use voice guide for rights questions, then benchmark the final API shortlist with the TTS evaluation protocol.
A two-path TTS and speech-to-speech production test
- Select Marketplace or Voice Lab; do not mix evidence across them.
- Obtain written consent before uploading a person’s voice, with project, media, territory, term and revocation scope.
- Build one TTS script and one STS performance containing names, numbers, emotion, whispers, shouts and timing constraints.
- Upload clean and deliberately imperfect source takes; record calibration and every setting.
- Compare at least two plausible target voices and accents, keeping identity/licensing evidence for each.
- Blind-grade intelligibility, identity similarity, artefacts, timing, emotion and localization with two reviewers.
- Request one late line change and one 25%/50% regeneration scenario.
- Reconcile submitted characters, source minutes, processed minutes, retries and accepted master minutes.
- Test API queues against slow/fast limits and verify 429 backoff without duplicate conversions.
- Test WAV download, project ZIP, starred-only export and ingestion into the real edit/master chain.
- If relevant, test the beta Pro Tools plugin on the supported production setup and preserve a fallback path.
- Delete dummy recordings/projects and request written retention, backup, model and log coverage.
- Obtain assurance, data processing, subprocessors, incident, support and SLA terms for the chosen offer.
- Calculate cost per accepted minute including performer/consent, engineering, editor and reviewer time.
Pass only when the exact voice and workflow meet quality, rights, throughput and deletion gates. A catalogue voice that sounds convincing but lacks a usable license fails; a fully licensed voice that requires too many corrections also fails.
Official sources