Audio prep and cost calculator

What it does

Open a recording and the tool:

• counts its length from the file's own bytes, before anything is decoded;
• converts it to mono 16-bit WAV at the sample rate you choose, from 8 kHz to 48 kHz;
• splits it into files no larger than a size you set, each a complete WAV file you can download;
• shows what it would cost to transcribe or send it, by each provider's price list, as far as the list could be read.

With no file, type a duration to see the sizes and the cost. The recording is decoded in your browser. The mixing and writing run in a Web Worker, and nothing is uploaded.

Try it: a one-hour meeting at 16 kHz, a generated five-second stereo tone, split at 0.1 MB, or thirty minutes at a rate of your own. The rate in that last example, $0.01 a minute, is illustrative and not any provider's price.

How it works

Which files. WAV (PCM, float, A-law, µ-law), AIFF, MP3, FLAC, and Ogg Vorbis or Opus. Each is decoded only if your browser can decode it: Chromium, for one, does not decode AIFF, and the tool says so when it cannot. Other files are refused: M4A/AAC, WebM (which is what a browser's own recorder writes), compressed WAV, and chained or multiplexed Ogg. The tool cannot count the length of those from their bytes, so nothing would limit the memory their decode takes. To keep one, convert it to one of the formats above first, or type its duration to see the sizes and the cost.

Its length is counted from the file's bytes, in a Web Worker, before anything is decoded:

• WAV and AIFF: the audio data to the end of the file.
• MP3: every frame header, at every byte, so that no header can hide the frames after it. A Xing or VBRI total is not read.
• FLAC: every frame header, checked by its checksum. The total in its STREAMINFO block is not read.
• Ogg: every packet, each at the most samples it can decode to, and the last granule position.

A header can state any length. The frames are what a decoder decodes, so they are what is counted.

Decoding happens once, when you convert. It uses the Web Audio API (decodeAudioData on an OfflineAudioContext at the rate you choose), and the browser resamples as it decodes. Mono is the average of the channels, the same down-mix the Web Audio API applies to stereo. WAV is 16-bit PCM: a 44-byte header and two bytes a sample, so an hour at 16 kHz is 115.2 MB.

Splitting cuts the samples into files of at most your size limit, header included, each as full as the limit allows. The cuts fall at the limit, not at a pause, and nothing listens for silence. A MB here is 1,000,000 bytes. Set your provider's own limit, because no provider's limit is assumed.

Only WAV is written. A compressed format such as MP3, Opus or FLAC would need an encoder the browser does not have, so the tool writes WAV.

The cost

Audio prices are not in Salyro's price list: Salyro does not carry audio. Each provider's price is a reference fact, recorded only after its own page has been read twice with the figures agreeing. When this tool was built:

• OpenAI, Google Gemini and xAI: the price pages could not be read, so their audio prices are unknown, not taken from memory.
• Anthropic: the pricing page was read and lists no audio price.

If you know your provider's rate per minute, type it. The tool applies it to the duration to the millisecond and labels the result as yours. A provider that rounds a recording up to the second or the minute bills a little more.

What it does not do

It does not transcribe, detect speech or silence, or send the audio anywhere. A shared link holds the settings, and a duration only if you typed one — never the recording or anything read from it.

Memory. A browser decodes the whole recording into memory before anything can look at it, often at the file's own rate: 43 minutes of stereo at 48 kHz is close to a gigabyte of samples, however small the file. The file size says little about that, and a file's own header can understate its length, so the tool counts the length from the frames first (above). It decodes nothing until you convert, and decodes once. The limit is 512 MiB for everything a conversion holds at its peak: the recording decoded at its own rate, the copy at the rate you chose (both for as many channels as it has), and the mono samples and the WAV written from them. That is about 14 minutes of 48 kHz stereo converted to 16 kHz, or about 40 minutes of 16 kHz mono. A longer recording is not decoded; type its duration to see the sizes and the cost. Files are limited to 25 MB, and the WAV it writes to 512 MB.

FAQ

Which sample rate should I choose?

The one your provider recommends; this tool does not assume one. A lower rate makes a smaller file: a minute of mono 16-bit WAV is 0.96 MB at 8 kHz, 1.92 MB at 16 kHz and 5.76 MB at 48 kHz.

Why is the WAV bigger than my MP3?

WAV is uncompressed: two bytes for every sample, where MP3 stores a compressed version of the sound.

Sources

The Web Audio API (W3C) for decoding and down-mixing, and the RIFF/WAVE format for the files written. Prices: Anthropic's pricing page, read; OpenAI's, Google's and xAI's price pages were tried and could not be read, and each fact on the page names the address that was tried.