Turning text into an MP3: built-in Windows voices, free tools and subtitle dubbing

Three routes, and how to choose

Turning written text into speech is TTS (text to speech). It used to sound unmistakably synthetic; neural voices have moved short sentences close enough to a human reading that most listeners don't notice. The catch is that those good voices are usually produced on a server somewhere, so the choice is mostly a trade between how it sounds and what it requires.

A WAV from the voices already in Windows

Windows ships with speech synthesis. Narrator and Edge's Read Aloud use it, but neither offers a save button. PowerShell can drive the same engine and write a WAV file. Open PowerShell from the Start menu and paste this:

Add-Type -AssemblyName System.Speech
$s = New-Object System.Speech.Synthesis.SpeechSynthesizer

# Which voices are available
$s.GetInstalledVoices() | ForEach-Object { $_.VoiceInfo.Name + " (" + $_.VoiceInfo.Culture + ")" }

# Pick a voice and a rate (-10 to 10, 0 is normal)
$s.SelectVoice("Microsoft Zira Desktop")
$s.Rate = 0

# Write it to a file
$s.SetOutputToWaveFile("C:\temp\narration.wav")
$s.Speak("Hello. Today we are turning text into speech.")
$s.Dispose()

An English install usually has Microsoft David Desktop and Microsoft Zira Desktop; other language packs add their own. For a long script, read the file and hand the whole thing over, so one script file becomes one audio file:

Add-Type -AssemblyName System.Speech
$s = New-Object System.Speech.Synthesis.SpeechSynthesizer
$s.SetOutputToWaveFile("C:\temp\narration.wav")
$s.Speak((Get-Content "C:\temp\script.txt" -Raw -Encoding UTF8))
$s.Dispose()

• The output is a WAV, which is large. Convert it if you need an MP3.
• Edge's Read Aloud uses much better voices, but there is no way to save what it plays.
• $s.Rate runs from -10 to 10. Around 0 to 2 is comfortable for instructions.

A natural voice in an MP3: free tools

The voices Microsoft Edge reads pages with are neural voices, and they sound far better. Free tools exist that save that audio to a file. Voxsave, built for this site, is one: paste a script or drop a subtitle file on it and it writes an MP3.

1. 1Install and run it, then paste your script or drop a subtitle file (.srt / .vtt) onto the window.
2. 2Pick a voice. Korean has a female (SunHi) and a male (InJoon) voice, and there are multilingual voices too.
3. 3Set the rate and pitch, and check it with the preview.
4. 4Save, and you get an MP3.

• No account, no login, no API key.
• The preview works offline, but saving an MP3 needs an internet connection, because Microsoft's servers do the synthesis.
• Windows 10 (1803 or later) and 11 only.
• Being unsigned freeware, it trips SmartScreen the first time: More info > Run anyway.

Automating it: edge-tts

The same voices are available from the command line. With Python installed, edge-tts lets you batch a folder of scripts or wire speech into a larger pipeline.

pip install edge-tts

# Text to MP3, with a matching .srt
edge-tts --text "Hello, world!" --write-media hello.mp3 --write-subtitles hello.srt

# List the voices
edge-tts --list-voices

# Read a script file with a chosen voice
edge-tts --voice en-US-AriaNeural --file script.txt --write-media narration.mp3

# Rate, volume, pitch
edge-tts --rate=-50% --text "Hello, world!" --write-media slower.mp3
edge-tts --volume=-50% --text "Hello, world!" --write-media quieter.mp3
edge-tts --pitch=-50Hz --text "Hello, world!" --write-media lower.mp3

Those are the options from the project's own documentation and command-line help. Loop over a folder and you can produce a series in one go. If you prefer a window to a terminal, the previous section is the same engine with a GUI.

Dubbing subtitles: the lengths never match

A subtitle file states exactly how long each line stays on screen. Read that line out loud and it almost always takes longer: people skim subtitles, while a synthesized voice pronounces every syllable. Translate the subtitles into another language first and the gap gets wider.

• Raise the rate. 10-20% faster still sounds natural.
• Shorten the line. The narration doesn't have to match the subtitle word for word; the same meaning in fewer words fits.
• Merge cues. Two short subtitles as one sentence give you room either side.
• If it has to be exact, cut the audio per cue in a video editor. That's faster than fighting the timings.

Cloud services and commercial use

Paid services offer more voices and finer control over delivery. In exchange, the free tier has a quota, and what you may do with the audio depends on the plan. If money is involved in the content, read that part first.

One thing sits apart from the plan you pay for: cloning a real person's voice without their consent is a problem whatever the service allows. Sticking to the stock voices avoids the question entirely.

FAQ