ReadSpeaker is a portfolio, not a single interchangeable TTS subscription. A content team may need speechMaker Studio for downloadable files; an app may need speechCloud API for streaming; a regulated device may need speechServer or an embedded/offline SDK. The wrong choice can produce a technically working voice with the wrong distribution rights or deployment model.
> Distinctive strength: ReadSpeaker offers separate browser authoring, cloud API, server and embedded/offline routes for accessibility and application delivery. > > Where it stops being an advantage: The visible product with the lowest friction may have the wrong deployment or distribution licence for the buyer's job.
This matters because the cheapest visible ReadSpeaker product may solve an unrelated job. TextAid is an education/reading tool, speechMaker Studio is browser authoring, speechCloud is a connected-app API, and licensed server/embedded products address on-premise or device use. Define the output and where it runs before asking for a price.
BenPicks has not tested ReadSpeaker voices, accessibility outcomes or implementation hands-on. This review maps current official product, pricing, terms, privacy, support and specification material into a procurement test. Vendor claims about naturalness, deterministic output or ease of integration still require evidence from your content.
The ReadSpeaker decision in 60 seconds
| Need | Product path | Material buying boundary |
|---|
| Produce downloadable narration in a browser | speechMaker Studio | Annual characters; one credit per downloaded character; five users on listed plans |
| Stream speech inside a connected app/device | speechCloud API | Public page says streaming only; static production/distribution needs separate license |
| Read websites/documents or serve learners | webReader/docReader/TextAid/LMS products | Priced by page views, enrolment/users or institutional agreement—not Studio capacity |
| Run TTS on-premise or centrally | speechServer/MRCP/SAPI | Licensed/quote-based server deployment |
| Run without cloud connectivity | speechEngine Embedded SDK | Android, iOS and Embedded Linux; other portability on request |
| Create a branded voice | VoiceLab/custom engagement | Consent, recordings, scope, deployment and rights require written agreement |
Which ReadSpeaker product should you buy?
Choose from the recurring deliverable:
- A team exporting voice files: speechMaker Studio.
- An app playing generated speech as a service: speechCloud API.
- A website or document accessibility layer: webReader/docReader.
- A learner support environment: TextAid or education integrations.
- A regulated, air-gapped or intermittently connected device: speechServer or embedded SDK.
- A unique brand/character voice: VoiceLab plus the intended runtime.
Do not accept a proposal named only “ReadSpeaker.” Require the exact product, hosting model, voices/languages, capacity metric, users, rights and support terms.
What does speechMaker Studio include?
Current Studio plans publish capacity rather than currency prices:
| Plan | Annual characters | Approximate words | Vendor-estimated speech hours |
|---|
| A | 270,000 | 54,000 | 5 |
| B | 540,000 | 108,000 | 10 |
| C | 1,350,000 | 270,000 | 25 |
| D | 2,700,000 | 540,000 | 50 |
All four are described as including five accounts, premium voices and full features. Larger annual tiers and custom voice support are sales-led. The page advertises 300+ Neural Standard/Premium voices in 90+ languages, browser projects, pronunciation/pacing/pitch/emphasis controls and SSML/IPA.
Treat the hours as planning approximations, not guaranteed output. Language, speaking rate, pauses and rejected generations alter accepted duration.
How are Studio credits counted?
Studio terms say one credit is charged for each downloaded character, excluding whitespace and including changes created by pronunciation-dictionary entries. Previewing does not consume credits. Purchased credits expire after 12 months and are non-refundable, including unused expired capacity.
That makes annual planning important. A Plan C workload that genuinely needs 1.35M final characters requires 1.6875M with 25% regeneration or 2.025M with 50%. Preview generously, but track every downloaded revision and do not buy annual capacity that cannot be scheduled before expiry.
The terms also say each conversion limit is set by the order confirmation and cannot exceed 20,000 characters. Long courses/books require segmentation and reliable assembly.
Can trial audio be used commercially?
No. Studio terms allow limited trial preview but disable download and prohibit commercial use of trial audio. The speechMaker Classic public demo likewise states evaluation-only use.
Paid Studio is designed for exported production audio, but retain the order confirmation and applicable terms. Confirm allowed channels, client work, redistribution, sublicensing and any voice-specific restrictions. A browser preview is not a license artifact.
Is speechCloud API licensed for downloaded files?
Not by default according to the public product page. speechCloud is described as streaming TTS for connected apps/devices, and the trial note explicitly says no static audio files may be produced, downloaded or distributed. Audio file production requires separate contact/licensing.
This is the biggest buying trap in the suite. If your application caches files, distributes podcasts, embeds audio in a course or stores generated prompts, put that use in the written entitlement. Do not infer file rights from the fact that the HTTP response contains audio bytes.
The API lists customer dictionaries, statistics/timing and optional SSML for pauses, phonemes and voice/language switching. Formats include A-law, u-law, PCM, WAV, Ogg and MP3 with format-specific rates. The retained public material did not establish a current universal table for requests, characters, concurrency or rate limits; obtain it for the exact account.
When does embedded or on-premise TTS make sense?
ReadSpeaker offers an embedded SDK for Android, iOS and Embedded Linux, with other platforms by request, plus speechServer/on-premise and hybrid deployment. Automotive and embedded pages describe offline operation for tunnels, remote areas, appliances, kiosks, robotics and industrial systems.
Offline processing can reduce network latency and keep text on device. ReadSpeaker states offline engines collect no user data. That statement belongs to the offline runtime; do not generalize it to cloud account portals, updates, telemetry or VoiceLab without a data-flow diagram.
Benchmark target hardware. Measure binary/model footprint, cold start, peak memory, synthesis latency, concurrent prompts, pronunciation updates, voice deployment, licensing checks and behaviour when storage or CPU is constrained. “Small footprint” is not a device result.
Are security and privacy documented?
ReadSpeaker publishes ISO/IEC 27001:2022 certification and an Intertek certificate. Its current privacy notice describes ReadSpeaker AB, GDPR handling and says personal/customer data or proprietary source code is not used to train third-party general-purpose AI unless otherwise noted.
Those facts are useful but not a complete product-specific answer. The notice does not establish one numeric retention schedule for every Studio script/audio/project, speechCloud request, custom voice recording, log and backup. Nor does a corporate ISO certificate automatically prove scope for every affiliate, cloud region and product.
Request the certificate scope, Statement of Applicability if available, architecture/data-flow, subprocessors, DPA, encryption, access logging, SSO/SCIM, deletion/backups and incident commitments for the chosen product.
What support should buyers expect?
The support page publishes a response target within 24 hours on business days for SaaS products and regional contacts across Europe, the Americas, Japan and South Korea. Licensed products have separate channels.
That is not the same as an uptime SLA. For critical IVR, safety announcements or embedded systems, require severity definitions, 24/7 coverage, response/restoration objectives, maintenance notices, service credits, offline fallback and version-support policy in the order.
What should you verify for a custom voice?
ReadSpeaker markets custom neural voices deployable in cloud, SDK and apps. The retained public material did not establish a complete end-to-end identity/consent procedure: who verifies the speaker, how scope and compensation are recorded, how revocation works, and when recordings/models/backups are deleted.
Before recording, obtain signed consent naming voice owner, project, languages, territories, channels, duration, model improvement, sublicensing, prohibited uses, withdrawal and post-termination treatment. Separate actor consent from your technical deletion test.
Who should shortlist ReadSpeaker?
Shortlist it when you need a specific enterprise TTS architecture: team authoring, accessible web/document reading, an API, on-premise server or offline embedded speech. The breadth is valuable when the same voice identity must span device, cloud and content production under a negotiated agreement.
Look elsewhere when you want one transparent self-serve price before specifying deployment, or a simple creator tool with public limits and checkout. Also pause if static-file rights, retention, custom-voice consent or SLA are unresolved.
Browse AI Voice software, read the commercial-use voice guide, and follow the TTS API benchmarking guide.
A fair ReadSpeaker evaluation protocol
- Name the exact product and deployment; reject a suite-level quote.
- Obtain a written entitlement schedule: voices, languages, capacity, users, streaming/static rights, channels and support.
- Use a rights-cleared multilingual script with names, numbers, abbreviations, phonemes and required emotional/pacing cases.
- Blind-grade exact candidate voices with native reviewers; record pronunciation corrections and rejected output.
- For Studio, track preview versus downloaded characters and model 25%/50% regeneration plus 12-month expiry.
- Test MP3/WAV/Ogg/PCM output in the actual editor/player and verify segmentation beyond the account's conversion limit.
- For speechCloud, test latency, errors, concurrency, statistics/timing, SSML, dictionaries and caching/file rights.
- For embedded/server, test target hardware, offline start, footprint, memory, throughput, update and failure behaviour.
- Map every script/audio/request/voice asset to storage, region, retention, logs, backup and deletion.
- If creating a voice, complete consent and technical revocation/deletion gates before any production recording.
- Verify ISO scope, DPA/subprocessors, access controls, incident process and binding SLA/support.
- Calculate total cost per accepted hour: license/quote + unused expiry + regeneration + engineering + reviewer time.
- Require a second reviewer to reproduce the product choice from retained evidence.
Pass only if the exact ReadSpeaker product meets content quality, distribution rights, architecture, capacity and governance gates. Suite breadth is not evidence that one contract includes everything.
Official sources