Voiceover and Licensing: Two Approaches—TTS and Recording It Yourself
There are two paths for AI short drama voiceover: use edge-tts to convert text into speech, or record it yourself on a phone. This page walks you through both paths, checks the licensing boundaries before publishing, and produces audio files ready to go straight into editing.
What You'll Be Able to Do After This Lesson
- Use edge-tts to convert a dialogue script into an MP3 audio file while also getting an SRT subtitle file
- Pick a suitable Chinese voice from the
--list-voicesoutput, and adjust the rate, volume, and pitch - Record clean, real-human voiceover sentence by sentence on a phone
- Clearly understand the licensing boundaries of TTS voices and know what to verify before publishing
Your Current Struggle
- Your short drama visuals are already generated, but there's no sound—the final cut doesn't feel like a "drama."
- You don't dare to record your own voice, or you don't have a quiet recording environment.
- You've heard of AI voiceover but worry whether the generated voices can be used commercially, fearing takedowns after publishing.
When to Use This Technique
When you need to: add spoken dialogue to your already-edited short drama visuals, and get a subtitle file at the same time.
And you don't want to: pay for a voice actor, or learn complicated audio software.
🎒 Preparation Before You Start
- Completed Trimming Dialogue: Subtract Before Voiceover, with a trimmed dialogue script in hand
- Completed Final Cut: Voiceover, Subtitles, Editing, Export, knowing where this page fits in the overall finishing workflow
- Approach A: a computer with Python installed and internet access (edge-tts calls Microsoft Edge's online TTS service)
- Approach B: a phone, plus a quiet room with little reverb
Core Idea
Both approaches produce the same thing: an audio file matching the dialogue script, ready for editing in the next lesson.
| Approach A: edge-tts | Approach B: Record on Your Phone | |
|---|---|---|
| Barrier to entry | Requires a computer and the command line | Just a phone |
| Feel | Standard TTS machine feel | Real-human feel, but depends on environment and equipment |
| Subtitles | The command produces SRT at the same time | Needs to be done separately in the next lesson |
| Best for | Lots of dialogue, fast iteration | Few lines, emotion needed |
Either works during the practice stage; before an official release, Approach A needs an extra licensing check (see Step 3).
Follow Along
Step 1: Generate Voiceover and Subtitles with edge-tts (Approach A)
First install the tool, then use a single command to produce audio and subtitles at the same time.
Install edge-tts:
pip install edge-ttsYou should see: the command finishes normally with no errors. (You can also install with pipx.)
Then list all available voices and find a Chinese one:
edge-tts --list-voicesYou should see: a voice list with columns such as name, gender, and language. Filter out the Chinese voices:
edge-tts --list-voices | grep zh-Voice names should be based on this command's actual output—don't copy old names from online tutorials, as Microsoft updates the voice list. Pick one whose gender and style match the character, and copy its name.
Generate the voiceover:
edge-tts --text "你终于回来了。" --voice <你选中的音色名> --write-media out.mp3 --write-subtitles out.srtReplace <你选中的音色名> with the name you copied in the previous step. You should see: two files appear in the current directory, out.mp3 (audio) and out.srt (subtitles). Keep the SRT subtitles for the next lesson.
Not happy with the rate or pitch? Add parameters and rerun:
edge-tts --text "你终于回来了。" --voice <你选中的音色名> --rate=-10% --pitch=-5Hz --write-media out.mp3 --write-subtitles out.srtNote that negative values must be written together (--rate=-10%, not --rate -10%). Volume uses --volume, with the same notation. Regenerate after each change, listen once, then finalize.
- Both
out.mp3andout.srthave been generated; listen to confirm the voice and rate are suitable
💡 Instant Playback
If you want to listen as it generates, you can use the edge-playback command for instant playback, but it depends on the mpv player (except on Windows). No mpv? No problem—just play the generated mp3 file directly.
Step 2: Record It Yourself on a Phone (Approach B)
When you don't have a computer or want a real-human feel, record sentence by sentence on your phone. The benefit of sentence-by-sentence recording: one take per line, so a mistake only means re-recording that one line, and it's easier to align during editing.
Steps:
- Open your phone's built-in recording app, choose the lowest-noise setting, and turn off air conditioning and fans.
- Keep your mouth about a fist's distance from the phone to avoid plosives.
- Record line by line following the dialogue script, pausing two seconds between lines for easier trimming later.
- Re-record any mis-spoken line in full—don't keep redoing takes within a single recording.
You should see: a set of audio files numbered by line in the recording app (or one long recording to be split later).
- Every dialogue line has a corresponding recording; playback shows no obvious noise or plosives
Step 3: Check Licensing Boundaries
Before publishing, confirm the licensing of the voice source. You can relax this during the practice stage, but not for an official release.
First, remember one distinction:
- Model license ≠ voice license. An open-source model doesn't mean the voices it generates can be used commercially at will.
- edge-tts calls Microsoft Edge's online TTS service, and its commercial terms are not explained in the edge-tts project documentation.@@I18N_KEEP_000006@@
- Your own voice belongs to you; Approach B has no third-party licensing issues (but music and sound effect licenses still need to be checked separately).
So:
Practice only, not published publicly: both approaches are safe to use.
Official release (especially commercial): Approach B is safest; for Approach A, you must verify Microsoft's terms of service yourself before publishing—this tutorial offers no legal guarantee.
Confirmed your publishing scenario (practice/public/commercial) and chosen a voiceover approach accordingly
Checkpoint ✅
- You have an audio file matching the dialogue script (mp3 or a phone recording)
- Approach A users: you also have the SRT subtitle file
- The voice name comes from the actual output of
--list-voices, not copied from a tutorial - You've clarified the publishing scenario and checked the corresponding voice licensing boundaries
Pitfall Warnings
❌ Hardcoding and Copying the Voice Name
Symptom: --voice reports the voice doesn't exist.
Cause: Voice names in online tutorials may be outdated—Microsoft updates the voice list.
Correct approach: Always run edge-tts --list-voices | grep zh- first and go by the actual output.
❌ Writing Negative Parameter Values Separately
Symptom: --rate -10% produces no change in the result, or a parsing error.
Cause: edge-tts requires negative values to be written with = attached.
Correct approach: write it as --rate=-10%, --volume=-50%, --pitch=-50Hz.
❌ Directly using TTS voiceover for commercial release
Symptom: the video is taken down by the platform or you receive complaints after publishing.
Cause: edge-tts's commercial terms are not stated in the project documentation, so assuming it is "commercially usable" by default has no basis.
Correct approach: before an official release, verify Microsoft's terms of service yourself, or use your own recordings instead.
Advanced Tips
If you want to drive character lip-sync with speech, you can look into S2V-14B (a speech-driven video model): just input --audio speech and --image reference images to generate a lip-synced video, and you can use --num_clip to control the number of clips for a quick preview. It has very high VRAM requirements (the single-GPU command is labeled 80GB), so it is an optional advanced route and is not required for this course.
Lesson Summary
✅ You have completed two approaches to AI short-drama voiceover:
- Using a single edge-tts command to get both MP3 audio and an SRT subtitle file
- Voice, rate, and pitch all adjusted based on
--list-voicesactual output and the negative-value concatenation rule - Sentence-by-sentence recording on a phone as an alternative with a real-human feel
- Distinguished model licenses from voice authorization, and verify the terms yourself before an official release
| Parameter | Recommended usage | Description |
|---|---|---|
--text | The condensed dialogue text | comes from the Dialogue Condensing lesson |
--voice | --list-voices The name in the output | do not hardcode it; go by the actual output |
--write-media | out.mp3 | Audio output |
--write-subtitles | out.srt | Subtitle output, used directly in the next lesson |
--rate / --volume / --pitch | --rate=-10%-style concatenation | negative values must be concatenated with = |
Next Lesson Preview
Once the voiceover files are in place, the next lesson Subtitle Creation and Proofreading: SRT from Generation to Burn-in covers subtitles. You will learn:
- How to proofread the timeline and typos in the SRT generated by edge-tts
- How to adjust subtitle styles so they don't block the picture
- How to burn subtitles into the final video