Smart Cache: pay only for the sentences you changed
07-09-2025 , 09-09-2026
Every sentence you voice is remembered for 7 days. Edit the text and convert again — SpeechGen re-voices only the sentences you touched and charges only for those. The rest comes back from the cache, free.
What it costs, measured
A script of five sentences, about 230 characters, voiced by a Pro voice where one credit equals one character. The same text was converted eight times in a row; these are the numbers the editor showed. The screenshots below come from such runs — exact figures shift a little with the wording, the proportions do not.
The first render is charged in full
Nothing of this text has been voiced before, so every sentence is new. There is no cache badge on the card — you paid for every character.
Change one sentence
A client asks for a correction in a single line. The other four sentences stay exactly as they were — character for character, which is what the cache matches on.
Convert again — pay for that sentence only
① The card is charged about a quarter of the first render. ② The badge reports how much of the audio came straight from the cache.
Both renders sit next to each other in the result list, so the difference is easy to see:
Add a sentence — pay for the new one
Appending a line to a finished script does not re-voice the script. Only the new sentence is synthesised. Here three words were typed at the end and nothing else was touched:
Some changes are free
Pause length and output format do not change the voiced audio itself — they change how it is assembled and encoded. ① a longer pause between sentences, ② a different file format:
Convert with both changes applied: zero credits, the whole text served from the cache.
Background music behaves the same way. It is mixed over the finished speech, so the voicing is untouched:
Converted again with the track playing under the voice — still nothing to pay, still a full cache hit.
This is why tuning pauses is safe to iterate on: try five variants of the pacing, pay nothing extra for any of them — and the same goes for trying background tracks under a finished script.
Speed and pitch belong to the voice
The speed and pitch controls sit on the voice row, not in the output settings. Moving either one asks the engine for a different performance, so every sentence is voiced again:
Convert and the cache badge is gone — the render is billed at full price, exactly like a first render.
Set the speed back to the value you used before and the next convert is free again: that version is still in the cache.
The free way to change speed or pitch
If you only want the finished track faster, slower or higher, process the audio instead of re-voicing it. Open Add effect on the result row and pick one — Slowed + Reverb, Echo, 8-Bit and the rest work on the file you already paid for, and Browse all effects opens the full catalogue:
Effects are software processing, so they spend no credits at all. They run against a separate daily processing allowance, and the voicing you already bought stays untouched.
The same effects work outside the editor too: the audio effects catalogue takes any file you upload from your computer, not only a voice-over made here. Speed, pitch, echo, reverb and the rest are applied the same free way.
Rule of thumb: change the voice row when you want a different performance, use an effect when you want the same performance treated differently.
The voice is the other full-price change
The cache key includes the voice. Pick a different speaker and the whole text is genuinely new audio — the badge disappears and the render is billed in full, just like a speed or pitch change.
Test voices on one short paragraph, then voice the full script with the winner. Switching back to a voice you already used is free again — its audio is still in the cache.
Free or charged — the full list
Anything that changes the sound of a sentence is re-voiced. Anything that only changes assembly or packaging is free.
Free — served from the cache
Re-converting an unchanged text · pause between sentences and paragraphs · output format and sample rate · background music and its volume · effects applied to a finished track · everything you did not edit in a long script · returning to a voice, speed or pitch you already used
Charged — new audio
Any edit inside a sentence, including punctuation and SSML tags · a new sentence · a different voice · a different speed or pitch on the voice row · a sentence that fell out of the 7-day cache
Where it saves the most
The longer the text and the more rounds of edits it goes through, the bigger the gap between paying per edit and paying per render.
Twenty lessons, a new intro line added to each. Only the intros are voiced; the lessons themselves stay free.
Opening hours change twice a year. Re-voice the one line that mentions them, keep the rest of the menu untouched.
Proofreading catches typos chapter by chapter. Each fix costs one sentence, not one chapter.
Same script, ten pacing variants before approval. Pause tuning is free, so only the copy edits cost anything.
Limits worth knowing
- 7 days. A voiced sentence stays in the cache for a week after its last use. Come back later and that text is billed as new.
- Up to 100,000 characters per text for the regular mode. Longer scripts switch to the large-block mode, up to 2,000,000 characters.
- Exact match only. An extra space or a changed comma makes the sentence new — the cache never guesses at "close enough".
- Per account. Your cache is yours; it is not shared between accounts.
- History for 30 days. Finished files stay in your profile for a month, independently of the cache.