WAV, M4A, or MP3 for TTS Voiceovers: What to Export

Choose PCM WAV for the narration master you will edit, and create a compressed delivery copy when the recipient or platform asks for one. Keep the master. Murmur currently exports WAV and M4A; it does not have a native MP3 export option. If you need MP3, convert a copy of the WAV master in an audio editor that supports it. Changing a file extension does not convert the audio.

The decision starts with the destination. An editor adding music needs a useful source file. A reviewer may need a small file that plays on their phone. A publishing platform may specify a codec, channel layout, or bitrate. Ask for those requirements before generating a whole project. This guide gives calculated size examples and a delivery checklist, not a claim that any one export meets every platform standard.

Separate the editing master from the delivery copy

PCM WAV stores uncompressed samples. MP3 uses lossy compression, trading some audio information for a smaller file. M4A is a container extension, so the name alone is not a complete description of the audio inside it. Inspect the actual export when a recipient has strict requirements. A valid file that plays in your own player can still be the wrong deliverable for a particular production pipeline.

Preserve a clear chain of files: original generation, edited master, and delivery copy. If the client changes a line, return to the master or project, replace that section, and make a new delivery copy. Repeatedly opening and recompressing the last MP3 can accumulate losses. Naming the master clearly helps a collaborator avoid using the smaller review copy by mistake.

Estimate storage before a long narration job

For uncompressed PCM audio, calculate the audio payload as duration in seconds multiplied by sample rate, channel count, and bits per sample, divided by eight. For constant-bitrate compressed audio, multiply duration by bits per second and divide by eight. Divide bytes by one million for decimal MB. Headers and metadata add some overhead; variable-bitrate files will differ. These are arithmetic examples, not measured Murmur export sizes.

Download the calculated audio-size examples if you want the assumptions and byte counts in a spreadsheet. The configurations illustrate format math; they are not a list of export controls in Murmur. A full project also needs room for revisions, intermediate renders, backups, and any music or effects. Budget for several versions rather than treating one final file as the entire storage requirement.

Use the speech time calculator for a rough duration estimate before you have generated audio. Then replace that estimate with the actual duration of a representative section. Pauses, delivery speed, and revisions can move the final duration substantially. Storage planning becomes more useful once you stop treating a word-count estimate as a measured runtime.

Keep sample rate and channels intentional

A sample rate describes how often the waveform is sampled; it is different from a compressed bitrate. A larger number does not automatically mean a better result. Converting lower-rate source audio to a higher rate can satisfy an editing requirement, but it cannot recover information absent from the source. Write down the source settings and the requested destination settings instead of applying the largest available numbers to every export.

For narration alone, a single mono channel can be a sensible source. A final mix with music or spatial effects may need stereo. Do not confuse a mono narration stem with the complete soundtrack. If an editor requests stereo, ask whether they need the finished mix or just a voice file compatible with their timeline. Supplying the right component is more useful than duplicating a channel without understanding the request.

Listen for a changed pitch or speed after any conversion. Correct resampling preserves the intended duration; simply relabelling the same samples with a different rate does not. If several generated sections came from different models, check every join after combining them. One successful opening segment does not prove that all later segments were interpreted at the correct rate.

A practical Murmur export handoff

• Finish the script and listen to the generated section before exporting. Use the Kokoro Mac guide if the source-generation step is still unclear.
• Export a WAV master and keep it alongside the project. The current native format choices are WAV and M4A; an MP3 sample on a website does not imply native MP3 export.
• Open the exported file in the receiving editor or player. Confirm that the intended section, ordering, and final sentence are present.
• Create a separate compressed copy only when needed. If using an external converter, check its privacy behavior before giving it confidential client audio.
• Send the recipient the filename, duration, format, and revision number. Keep the previous approved version until the replacement has been accepted.

Murmur is designed for local speech generation on Apple Silicon Macs running macOS 15 or later. Kokoro is bundled for basic narration. Optional models need extra setup and resources. Local generation does not mean the app never connects to the internet: activation, periodic license validation, updates, model downloads, and other network features still apply. Review the project workflow before committing a large job to any new tool.

Use this final delivery checklist

Listen to the exported file itself, not just the in-app preview. Start at the beginning, inspect each edit boundary, and play the final sentence through its ending. Then check a section in the middle where numbers, names, or a speaker change occur. For high-value work, listen to the full file. A quick spot check is useful for catching obvious export problems, but it cannot certify every word.

• The filename identifies the project, section, and revision without exposing unnecessary private details.
• The file opens in the recipient's actual editor or player, and the duration matches the expected version.
• Words are complete, chapter order is correct, and joins do not introduce accidental gaps or overlap.
• Levels are comfortable and do not distort; a loudness requirement is checked separately from the file extension.
• The editable project and master remain available so a later correction does not start from a compressed review copy.

If the file sounds too quiet, diagnose levels separately from format choice. Converting WAV to MP3 does not inherently solve a quiet recording. The audio normalizer may help with level adjustments, but check the preview and exported result, especially around loud syllables. Avoid processing an entire archive until one representative file survives the intended workflow.

Questions about narration formats