Revising a SpeechGen voiceover: choose the voice, edit the sentences, then tune the pauses
For course or product-demo voiceovers, audition voices and speaking speeds with a short sample before consolidating text edits and adjusting pauses. SpeechGen’s sentence cache can reduce repeated synthesis, but it depends on matching text and voice settings. A file remaining in your history does not guarantee further revisions are free.
Once a course voiceover reaches review, feedback often includes both “this sentence is wrong” and “leave some time here to look at the screen.” In SpeechGen, these edits should not be handled the same way: the first requires a text change, while the second can start with a pause adjustment. Casually switching voices or changing the whole script’s speaking speed each round can turn a one-sentence fix into a full-script resynthesis.
A workflow suited to repeated proofreading settles the voice first, then keeps the wording and settings of approved sentences stable. The following product-demo narration example explains how the work fits together, based on official documentation rather than tests of audio quality or money saved.
Use a short sample to settle choices that affect the whole script
Choose a passage containing a product name, numbers and a long sentence. Select the correct language in the editor, preview candidate voices, then generate that passage. Have the reviewer confirm pronunciation, voice and speaking speed before adding the complete script.
This narrows the scope of trial and error. SpeechGen matches cached audio by sentence and voice settings; choosing a different voice or changing speed or pitch on the voice control row may require the original script to be synthesized again. Do not treat “just a little faster” as a cost-free formatting change. If you use SSML to control pronunciation, test the tags with the chosen voice, because the official documentation says not every voice supports every tag.
Record the selected voice and settings alongside the script version. When someone else takes over, check this information first so the same passage is not assigned a different set of voice parameters.
Attach review comments to specific sentences
After generating the full script, listen through it, then organize feedback by sentence: which was mispronounced, which needs rewriting and where only the pause is too short. Change only the sentences that need editing rather than tidying punctuation or spacing throughout the script at the same time.
The official cache documentation requires character-for-character matches. Even changing a comma makes its sentence new content; with other conditions unchanged and a valid cache, untouched sentences can be reused directly. Revision notes should therefore distinguish content changes from pacing adjustments instead of prompting a fresh polish of the entire script every round.
For example, if a feature name in the demo changes, edit and regenerate the relevant sentence. If the name is unchanged and viewers simply need more time to study the screen, try a pause first. This distinction does not replace listening, but it makes clear what each change is meant to solve.
After adjusting pauses, check the audio against the visuals
Open Settings below the text box and find Pause Control to adjust sentence and paragraph pauses separately. According to the documentation, changing only pauses does not require voice resynthesis when a valid cache is used. The paragraph pause is added to the pause after the paragraph’s final sentence, so do not judge total silence from a single dropdown value.
Global settings affect the entire script. Use them to establish a broad rhythm, then listen at paragraph boundaries, especially where the demo changes screens. Longer pauses may aid comprehension, but they also change the track’s total duration. Put the audio back on the editing timeline to check it; sounding smooth in the editor alone is not enough for delivery.
Download the audio once it is approved. The official guide says MP3 is available by default. For WAV or OPUS, choose the relevant output settings, then regenerate and download. Save the final audio, matching text and voice settings together at handoff so the video editor does not pair an old track with new visuals.
Cache lifetime and delivery archives are different things
SpeechGen’s sentence cache lasts 7 days from its last use and is not shared between accounts. Sentence-level processing in regular text mode applies to texts of no more than 100000 characters each. Longer scripts switch to large-block processing, so do not assume the same savings from sentence-level revisions.
Project history lasts 30 days, while the documentation recommends adding important files to favorites for long-term storage. Favoriting preserves the finished work, not the sentence cache indefinitely. If approval runs past cache expiry, reconsider usage before synthesizing again rather than assuming revisions are free because the old file is still downloadable.
This workflow is therefore better suited to short projects where feedback can be collected together. Settle the voice with a short sample, consolidate text corrections, adjust pauses and check against the visuals, then archive the result. If your team often returns much later to change the voice throughout, treat resynthesis as normal work rather than relying on the cache as a promise of free long-term revisions.
Content note:This article is based on publicly available information and is intended as a reference for evaluating tools. Features, pricing, and related policies may change at any time; please refer to the latest official information.
Original source:speechgen.io