Commercial-use eligibility has not been established.
Main caution
Resolve the open evidence fields before buying.
From evidence to action
Make the SpeechGen decision with the right unit and route.
Each module separates documented facts, calculations and editorial conclusions. Missing or incompatible evidence stays visible instead of becoming a guess.
The number that changes the decision
Cache-aware cost calculator
SpeechGen: Characters, voice tier and changed-sentence percentage estimate initial and revision cost.
pricing · no subscription
Credit packs are prepaid, have no auto-renewal and remain valid for one year; topping up resets the timer for the combined balance.
pricing · packs
€4.99/25k, €9.99/65k, €24.99/200k, €49.99/500k, €149.99/2M and €599.99/10M credits.
pricing · standard rate
One credit covers two characters.
pricing · pro rate
One credit covers one character.
pricing · hd rate
Two credits cover one character.
pricing · free
Pricing header displays 2,000 free credits, while policy text says eligible email domains receive 1,000 limits after activation.
Decision-changing number
Characters, voice tier and changed-sentence percentage estimate initial and revision cost.
Keep in mind: Use only the retained SpeechGen evidence; do not generalize this decision asset to another product.
Open the complete SpeechGen buying analysisWorkflow · pricing · evidence boundaries · evaluation · official sources
SpeechGen's strongest feature is not its advertised 5,000-voice catalogue. It is a pay-as-you-go production model built around long text and sentence-level caching: unchanged sentences remain reusable without another charge during a seven-day window, while credits can be mixed across Standard, Pro, HD voice and transcription.
That can materially reduce revision cost for audiobooks, courses, IVR libraries and serialized narration. It also creates a buying task: Standard, Pro and HD consume the same balance at very different rates; credits expire after a year unless topped up; clone storage spends credits every day; and refunds become narrow after payment. BenPicks has not tested SpeechGen hands-on, so quality, cache and workflow claims remain vendor claims until reproduced with a controlled project.
> Distinctive strength: Sentence-level caching can make a small correction genuinely cheaper than regenerating a whole chapter. > > Where it stops being an advantage: The saving depends on unchanged sentences staying inside the documented cache window; voice tiers, transcription and clone storage draw from the same balance at different rates.
SpeechGen is a revision-economics decision
Decision point
Source-led position
Strongest fit
Long-form creators and developers making iterative, exportable narration
Distinctive strength
Pay-as-you-go credits plus free reuse of unchanged cached sentences
Public packs
$4.99/25k to $599.99/10M credits
Voice-tier consumption
Standard: 2 chars/credit; Pro: 1; HD: 0.5
Long-text API
Up to 1M characters; browser FAQ says 2M per generation
Commercial use
Explicitly permitted, subject to input rights and content rules
Voice cloning
2,000 credits creation + 250/day storage; written consent required
Main unknowns
API rate limits, training use, enterprise security and SLA
Smart Cache, not catalogue size, is the economic edge
> Distinctive strength: SpeechGen combines pay-as-you-go credits with sentence-level caching, allowing unchanged sentences to be reused without another charge during the documented cache window.
SpeechGen combines a browser production editor with a practical long-text API. It can ingest pasted text, DOCX, PDF or subtitles; assign multiple voices; apply SSML, pauses, pitch and speed; split a generation into named files; and export several audio formats. The `/longtext` API accepts book-scale jobs and the `/subs` path adds timestamps.
The differentiator is how revisions are metered. SpeechGen says each sentence is cached for seven days; unchanged sentences reused in a regenerated project cost zero credits. A creator can correct one paragraph without paying again for an entire chapter. That is a real economic advantage over systems that meter every regenerated character.
The boundary is temporal and technical. The cache lasts seven days, changing speed or pitch requires full regeneration, and quality tiers spend credits differently. A project rewritten after the cache window receives no promised saving. A buyer must reproduce the balance change rather than assume every export is free.
Use the AI Voice software category to compare this hybrid editor/API with creator studios, localization systems and low-latency infrastructure.
One balance buys very different amounts of speech
Current public packs range from $4.99 for 25,000 credits to $149.99 for 2 million. A $599.99 10-million-credit tier is described as exclusive/invitation-oriented, even though its checkout page is reachable. Credits remain valid for one year; buying another pack adds the balance and restarts the timer.
The voice tier changes purchasing power:
Pack
Standard characters
Pro characters
HD characters
$49.99 / 500k credits
1,000,000
500,000
250,000
$149.99 / 2M credits
4,000,000
2,000,000
1,000,000
$599.99 / 10M credits
20,000,000
10,000,000
5,000,000
At the 2-million pack, one million characters cost about $37.50 Standard, $75 Pro or $150 HD. At 840 characters per minute—the page's own English approximation—that is roughly 1,190 Standard, 595 Pro or 298 HD minutes per million characters. Actual language, speed and punctuation change duration, but billing remains per character.
Add regenerated changed sentences, expired cache, transcription and clone storage. A single active clone costs 250 credits/day: 91,250/year, before any synthesis. That equals 182,500 Standard characters, 91,250 Pro or 45,625 HD characters of opportunity cost.
Use the AI voice cost guide to convert approved output rather than headline minutes.
Does Smart Cache really make revisions free?
Only unchanged cached sentences. SpeechGen says sentence audio remains cached for seven days and that regenerating a project charges only new or edited sentences. Pauses between sentences/paragraphs can be changed without consuming limits, while changing speed or pitch affects synthesis and consumes credits.
The safest test is mechanical:
record the account balance;
generate a ten-sentence fixture;
export it and record consumption;
edit exactly one sentence inside seven days;
regenerate and compare the debit;
repeat after changing speed;
repeat after the cache expires.
Do not call the workflow free unless account history proves the expected debit. Keep source text and finished audio outside SpeechGen; project history is not a permanent archive by default.
Can SpeechGen handle books and automated production?
The browser FAQ claims up to two million characters in one generation and supports up to 1,000 short or 500 long segments. The API exposes three routes:
`/text`: immediate audio, up to 2,000 characters;
`/longtext`: queued job/polling, up to one million characters;
`/subs`: queued audio plus timestamps.
The limits are not contradictory because they describe different interfaces. Long jobs return a task ID and require polling `/result`; production code must handle processing, success and error states and download the result before its URL disappears.
Authentication uses both account email and a secret token. The API documentation permits GET query parameters as well as POST. Do not put credentials into URLs: query strings can enter logs, browser history and monitoring. Use POST, server-side secrets and HTTPS. A general rate/concurrency or SLA contract was not located, so test controlled parallel jobs and ask support before committing a bulk deadline.
Long-form production needs more than an MP3 button
The API documents MP3, WAV, OGG, Opus and FLAC, with sample rate, bitrate and mono/stereo controls. Editor features include speed, pitch, emotional style and intensity, voice role, pauses, volume, effects, SSML and voice-dependent intonation graphs. Dialogue markup assigns voices to characters, and cut markers create separately named files.
Not every voice supports every style or SSML feature. “5,000+ voices” and “150+ languages” describe breadth, not usable combinations. Select the intended language, then check the precise voice tier, styles, sampling formats and credit multiplier. Blind-listen to Standard, Pro and HD before paying two or four times the character cost.
Commercial use comes with disclosure and content rules
SpeechGen's public policy says users may dispose of generated files without restriction, including commercial use, while remaining responsible for rights in input text. It also imposes content rules: synthetic nature must be disclosed so users are not deceived, a feedback channel is required, impersonation/deception and specified harmful uses are prohibited.
That is clearer than a vague sales-page licence. Still, businesses should retain the policy version, invoice and source rights. Confirm whether stock voices or third-party providers impose narrower rules for a particular voice, especially advertising, resale, political, health, financial or legal content.
SpeechGen makes clone consent unusually explicit
The current product page publishes three separate meters: 2,000 credits to create a clone, 250 credits/day while it remains active, and synthesis at the HD voice rate. It accepts up to three files, 25 MB each, and recommends 12–60 seconds of single-speaker audio. Fifteen languages are advertised: nine stable and six experimental.
The consent contract is unusually specific. A third-party speaker must give explicit, informed, documented written consent identifying SpeechGen, creation and generation, purpose and scope. Evidence must remain available during the clone's life and for two years after deletion. SpeechGen may demand it within seven days. Minors are prohibited, and political/public-safety impersonation categories are restricted.
Withdrawal requires immediate clone deletion. The vendor says deletion permanently removes the model and stops storage charges. Test that behavior and retain your consent evidence independently. Use the voice-cloning consent guide to verify revocation, disclosure and auditability.
What happens to projects and personal data?
The policy says registered project history is stored for 30 days and unregistered files for 24 hours. The FAQ says favorites are kept permanently; because “permanent” is not a backup SLA, export every approved file and project manifest.
The privacy policy provides withdrawal of personal-data consent and describes legal/technical protections, but it does not state a clear general rule for using ordinary customer scripts/audio in model training. Nor did this review locate public SOC 2, SSO, DPA, residency, private deployment or SLA documentation. Do not submit confidential or regulated material until those points are resolved contractually.
The refund window is narrow: request within 24 hours, with no more than 3,000 characters consumed. Unspent limits otherwise are non-refundable. Use the free allowance before purchasing and immediately run the decisive voice, format and rights tests.
SpeechGen suits iterative long-form work
Shortlist it when:
long-form narration benefits from sentence-level revision caching;
pay-as-you-go packs fit intermittent production better than subscriptions;
multiple formats, dialogue, segmentation or a long-text API matter;
commercial-use wording and synthetic disclosure fit the project;
transparent clone creation/storage meters are preferable to sales contact.
Look elsewhere when:
a contractual API SLA/concurrency guarantee is required publicly;
the project needs SOC 2, SSO, DPA, residency or private deployment evidence;
a subscription with predictable included premium minutes is simpler;
indefinite project storage must be part of the service;
the buyer cannot evaluate many third-party voices and tier multipliers.
For API-oriented alternatives, compare the current best AI voice APIs for developers with the same accepted-output and governance tests.
Prove the cache with a one-sentence edit
Use a 20-minute chapter containing dialogue, proper nouns, numbers, a language switch and SSML. Then:
record free/paid credit balance and exact tier consumption;
compare Standard, Pro and HD blind;
edit one sentence and verify Smart Cache debit;
export MP3 and lossless WAV/FLAC and inspect loudness;
split named segments and reassemble them;
submit through `/longtext`, handle polling/error and delete the project;
confirm history/favorite retention and external backups;
create/delete a consented test clone and verify daily charges;
preserve commercial rights and synthetic-disclosure workflow;
run limited concurrency and obtain missing security/SLA answers.
Reject the product if premium voices do not justify their multiplier, cache savings cannot be reproduced, production throughput is undocumented/insufficient or data/security answers do not fit the content.
Bottom line
SpeechGen has a credible long-form advantage: it combines pay-as-you-go credits, a production editor, a million-character API and sentence-level caching that can make targeted revisions materially cheaper. Its public cloning price and consent rules are also unusually concrete.
The catalogue size is not the reason to buy. The decision depends on accepted-output cost across Standard/Pro/HD, reproducible cache behavior, throughput and whether the missing enterprise/data commitments matter. Test those boundaries before a large credit purchase because the refund window closes quickly.
Make Smart Cache earn the purchase
Use the free allowance to generate a ten-sentence project, change one sentence and confirm the debit before buying credits. If the cache behaves as documented and the selected voice tier survives a full chapter, purchase only the pack that covers the next finished project plus measured corrections. If the buyer needs an SLA, private deployment or documented enterprise controls, the absence of those terms is a reason to compare providers—not to prepay a larger balance.
SpeechGen privacy and service policy ↗ — free evaluation, refund, retention, commercial rights and prohibited uses; source availability rechecked 2026-08-30.
