Almacenamiento de grabaciones de llamadas: tamaños de WAV frente a MP3
How much storage call recordings need: the bitrate maths behind WAV and MP3, a sizing formula you can reuse, and why retention and legal hold drive the total.
Respuesta rápida
Recording size equals bitrate times time. G.711 telephone audio is 8,000 samples a second at 8 bits, so 64 kbit/s: 8,000 bytes a second, 480 KB a minute, 28.8 MB an hour per channel. Total storage is that figure multiplied by recorded hours, channels and retention days. Format changes the result by a factor of two or four; retention and legal hold can change it by far more.
Published figures for how much storage call recordings need often disagree. The reason is usually not that anyone is wrong: each answer silently assumes a different bit depth, sample rate or channel count. Once you derive the number from the audio itself, the confusion goes away, and you can build a sizing sheet you trust.
This guide starts from the codec, shows the arithmetic, explains why WAV and MP3 are not a like-for-like comparison, and walks through a reusable formula with an illustrative worked example. It then explains why retention and legal hold usually matter more than the choice of format.
Where the bytes come from
An uncompressed digital audio stream has a size you can calculate:
bytes per second = sample rate x bits per sample x channels / 8 Telephone audio is narrow. RFC 3551 defines the two G.711 payloads, PCMU and PCMA, as audio with 8-bit samples and an 8,000 Hz clock rate, packetised at 20 ms by default (RFC 3551, section 4.5.14 and its payload table). ITU-T G.711 is the Recommendation titled "Pulse code modulation (PCM) of voice frequencies" that defines this encoding (ITU-T G.711). So one channel of G.711 gives:
- 8,000 samples per second x 8 bits = 64,000 bits per second = 64 kbit/s
- 64,000 / 8 = 8,000 bytes per second
- 8,000 x 60 = 480,000 bytes per minute (about 0.48 MB)
- 8,000 x 3,600 = 28,800,000 bytes per hour (about 28.8 MB)
These are decimal megabytes (1 MB = 1,000,000 bytes), which is how most storage is sold. Binary megabytes (MiB) give a smaller-looking number, about 27.5 MiB per hour. Pick one convention and keep it in the sheet.
The 20 ms packet interval is also useful: it means 50 packets per second, each carrying 160 bytes of audio (8,000 bytes / 50). That is the audio only. The headers on the network are not stored in a recording.
WAV is a container, not a codec
"WAV versus MP3" is a slightly unfair comparison. The Library of Congress describes WAVE as a wrapper file format that can incorporate an audio bitstream with other data chunks, and a subtype of RIFF. It notes that a common encoding inside it is linear PCM (LPCM), and lists mu-law and A-law as encodings it may contain (Library of Congress, Format Description for WAVE). So the size of a WAV depends on what is inside:
| What is stored | Rate | Per hour, mono |
|---|---|---|
| WAV with mu-law or A-law, 8 kHz, 8-bit | 64 kbit/s | 28.8 MB |
| WAV with linear PCM, 8 kHz, 16-bit | 128 kbit/s | 57.6 MB |
| WAV with linear PCM, 16 kHz, 16-bit (wideband) | 256 kbit/s | 115.2 MB |
This is the main reason published figures disagree. One source assumes the 8-bit G.711 stream (28.8 MB an hour); another assumes the 16-bit linear PCM that many recorders write (57.6 MB an hour). Both are right for their assumption. Before you trust any number, ask: what sample rate, what bit depth, how many channels?
A practical point about sample rate: a narrowband phone call carries 8 kHz audio. Saving it at a higher sample rate makes the file bigger without adding detail that was never in the call. Wideband audio is a different case, for example when the call uses a wideband codec. The Opus codec, defined in RFC 6716, supports bitrates from 6 kbit/s to 510 kbit/s (RFC 6716, section 2), so an Opus call could be well above or below G.711 in rate. Record at the rate your calls were actually carried.
MP3 and other compressed formats
A compressed format such as MP3 stores the audio at a bitrate you choose when the file is written. The arithmetic is the same: bytes per second equal bitrate divided by eight. The table below is plain arithmetic for three example bitrates; the right choice for you depends on a listening test, not on the table.
| Bitrate | Bytes per second | Per hour | Compared with G.711 at 64 kbit/s |
|---|---|---|---|
| 32 kbit/s | 4,000 | 14.4 MB | Half |
| 64 kbit/s | 8,000 | 28.8 MB | Igual |
| 128 kbit/s | 16,000 | 57.6 MB | Double |
The key trade-off is between size and what you need the recording for. Three uses pull in different directions:
- Quality assurance listening. A human reviewer is usually fine with moderate compression, provided speech stays clear. Test with real calls, including noisy ones, before settling.
- Transcription. Speech recognition can be sensitive to compression artefacts. Run the same sample set through each format and compare the word error rate; our guide on how to measure call transcription accuracy explains how.
- Disputes and evidence. If a recording may be used to settle a disagreement, fidelity and a clear chain of custody matter more than saving a few gigabytes. Where that applies, a higher-fidelity format is a reasonable choice. Ask your advisers what your situation needs.
A sensible approach is to choose by purpose per campaign: a compressed format for routine QA, a higher-fidelity one where fidelity is the point.
Mono versus dual-channel
A recording can mix both parties into one channel (mono) or keep them on two channels (dual, or stereo). Dual-channel doubles the uncompressed size: G.711 becomes 57.6 MB an hour for the pair. In return it keeps the agent and customer voices separate, which makes speaker attribution simple, helps transcripción avoid confusing who said what, and lets you measure talk-over and silence by side. If you do not need those, mono halves the bill. Decide this before you size, because it is a straight factor of two.
A reusable sizing formula
storage = agents x talk hours per agent per day x working days x retention x MB per hour x channels Keep these inputs in a sheet, each with a note on where it came from:
- Recorded hours per day. Use recorded talk time, not shift time. Calls that are never answered produce little or nothing to store, depending on what you record, so check whether you record ringing or only connected audio.
- MB per hour per channel. From the tables above.
- Canales. 1 or 2.
- Retention in days. The period before automatic deletion, plus a buffer for the cleanup schedule.
- Legal hold allowance. A separate line for recordings kept beyond normal retention.
- Overhead for backups and copies. If you keep a second copy, multiply.
Worked example (illustrative only)
Assumptions, invented for the arithmetic and not taken from any customer: 50 agents, 4 recorded talk hours per agent per day, 22 working days a month, G.711-rate audio at 28.8 MB an hour, mono, 12 months of retention.
- Recorded hours per day: 50 x 4 = 200 hours.
- Per day: 200 x 28.8 MB = 5,760 MB, about 5.76 GB.
- Per month (22 days): 5.76 GB x 22 = about 126.7 GB.
- 12 months of retention: about 1.52 TB.
Now change one input at a time:
| Change | Result | Efecto |
|---|---|---|
| Baseline (28.8 MB/h, mono, 12 months) | about 1.52 TB | 1.0x |
| Dual-channel | about 3.04 TB | 2.0x |
| 16-bit linear PCM at 8 kHz | about 3.04 TB | 2.0x |
| Compressed at 32 kbit/s (14.4 MB/h) | about 0.76 TB | 0.5x |
| Retention 36 months instead of 12 | about 4.56 TB | 3.0x |
| Retention 3 months instead of 12 | about 0.38 TB | 0.25x |
The format choices move the answer by a factor of two. The retention choice moves it by a factor of three or four, and every extra month of retention adds another month of recordings to the total, because nothing ages out until the retention period ends. This is why retention is the biggest lever.
Retention, legal hold and deletion
Retention is a policy decision, not a technical one. Different regions, industries and contracts ask for different periods, so decide with qualified advisers and write the answer down; this guide does not tell you how long you must keep recordings. See the guía de cumplimiento para la grabación de llamadas for the wider picture of consent and retention.
Three storage-sizing consequences follow:
- Growth is linear until the first deletion. In the first year of a 12-month policy, storage rises every month. After that it plateaus at the steady-state figure, as old recordings age out. Plan capacity for the plateau, not for month one.
- Legal hold breaks the plateau. A recording on hold is kept until the hold is released, so held volume accumulates outside the normal cycle. Give it its own allowance and review it regularly.
- Deletion must be verifiable. A retention job that fails quietly keeps data longer than your policy says. Check storage graphs against the expected plateau. A flat line that should be falling is a warning.
Cómo encaja DialerBee
El grabación de llamadas feature records automatically and supports both MP3 (storage efficient) and WAV (maximum audio fidelity), configurable per campaign or globally. Recordings are played back through time-limited signed URLs, retention is configurable per campaign and per tenant, and recordings on legal hold are preserved and excluded from automatic purge. Storage usage is monitored per tenant, campaign and time period, with alerts when usage approaches storage limits. That fits the approach above: pick the format per campaign, set retention per campaign, and watch the usage against the figure you calculated. For a wider budget view, the call centre cost calculator is a good companion, and the voice codec glossary entry explains the codecs that set the underlying bitrate.
Recording quality is also part of a good QA process; the outbound call quality assurance guide shows how reviewers use recordings.
A sizing checklist
- Write down sample rate, bit depth and channels for each campaign.
- Calculate MB per hour from the formula, not from a vendor chart.
- Use recorded talk hours, not shift hours.
- Choose mono or dual-channel by what you will do with the recording.
- Compare formats on real calls for QA listening and transcription.
- Agree retention periods with advisers, then size for the plateau.
- Add a separate allowance for legal hold and for backup copies.
- Set a storage alert and check the graph against your forecast every month.
Preguntas frecuentes
How big is one minute of call recording?
For uncompressed G.711 audio (8,000 samples per second at 8 bits each) one mono minute is 480,000 bytes, about 0.48 MB, which is 28.8 MB per hour. A 16-bit linear recording at the same 8 kHz rate is double that, about 57.6 MB per hour. A compressed MP3 depends on the bitrate you choose: bytes per second equal the bitrate divided by eight.
Is WAV always bigger than MP3?
Not by definition. WAV is a container, not a codec. It can hold linear PCM, mu-law or A-law audio. A WAV that holds 8-bit mu-law at 8 kHz is the same size as the original G.711 stream, while an MP3 encoded at a higher bitrate can be bigger. Compare the audio encoding and its bitrate, not the file extension.
Should I record in mono or dual-channel?
Dual-channel keeps the agent and the customer on separate channels, which helps transcription, speaker attribution and disputes, and roughly doubles the size of an uncompressed recording. Mono mixes both voices into one channel and is half the size. Decide by what you need to do with the recording afterwards.
What drives total storage more: format or retention?
Retention. Storage is volume per day multiplied by the number of days you keep recordings. Halving the file size halves the total, but doubling the retention period doubles it, and a legal hold can keep recordings far beyond the normal window. Model retention first, then pick the format.
Does lower bitrate hurt transcription accuracy?
It can. Heavy compression removes detail that speech recognition uses. Do not assume; run a small test. Transcribe the same sample calls at each format and compare word error rate, as described in our guide to measuring transcription accuracy.
Can I delete recordings on a schedule and still keep some?
Yes, if your platform supports both. In DialerBee, retention is configurable per campaign and per tenant, and recordings on legal hold are excluded from automatic deletion. Check which rules apply to you with qualified advisers before choosing a period.
Fuentes
- RFC 3551, RTP Profile for Audio and Video Conferences with Minimal Control (PCMU and PCMA: 8-bit samples, 8,000 Hz clock rate, 20 ms default packetisation). rfc-editor.org/rfc/rfc3551 (consultado el 7 de noviembre de 2026)
- ITU-T G.711, Pulse code modulation (PCM) of voice frequencies. itu.int/rec/T-REC-G.711 (consultado el 7 de noviembre de 2026)
- Library of Congress, WAVE Audio File Format (RIFF wrapper; LPCM, mu-law and A-law encodings). loc.gov/preservation/digital/formats/fdd/fdd000001.shtml (consultado el 7 de noviembre de 2026)
- RFC 6716, Definition of the Opus Audio Codec (6 to 510 kbit/s). rfc-editor.org/rfc/rfc6716 (consultado el 7 de noviembre de 2026)
This article is technical guidance. DialerBee does not provide legal advice or guarantee regulatory compliance. Retention periods and recording rules differ by region; check yours with qualified advisers.
Lecturas relacionadas
Artículos relacionados
Cómo elegir un proveedor de troncales SIP para un marcador
9 minutos de lectura
Tecnología¿El DTMF no funciona en una troncal SIP? Cómo corregir las pulsaciones de teclas
Lectura de 10 minutos
OperacionesEstrategia de reintentos de llamadas: reglas de remarcación por resultado
Lectura de 11 minutos
CumplimientoLlamadas de cobranza bajo la Ley 21.320 de Chile: una llamada por semana
Lectura de 10 minutos
¿Listo para ver DialerBee en acción?
Solicite una demostración en vivo de 15 minutos o comience una prueba gratuita de 14 días: sin presentaciones, sin compromiso. Las llamadas empiezan en cuanto su propia troncal SIP o PBX está conectada y aprobada.
Prueba gratuita de 14 días · crédito de IA gratuito · sin tarjeta · 11 idiomas · BYOC · controles que apoyan el cumplimiento