Play PCM File Online

PCM (pulse-code modulation) is the simplest way to store sound digitally. A microphone signal is measured thousands of times per second, and each measurement — a sample — is written down as a number. Play those numbers back at the same rate through a speaker and you get the original sound. Every WAV file, every CD and nearly every audio interface works with PCM internally; MP3, AAC and Opus are compressed formats that are decoded back to PCM before you hear them.

A raw .pcm file is just those numbers, one after another, with nothing else in the file. There is no header saying how many samples per second there are, how many bytes each sample occupies, or whether left and right channels are interleaved. Three facts fully describe the data — sample rate, bit depth (with its encoding and byte order), and channel count — and all three have to come from somewhere other than the file itself: the device documentation, the API that produced it, or trial and error.

PCM vs WAV: the 44-byte difference

A WAV file is PCM with a small RIFF header in front, usually 44 bytes long. That header stores exactly the three facts a raw file is missing, which is why any media player can open a WAV but almost none can open a .pcm. If you strip the header from a 16-bit WAV, what remains is a valid raw PCM file. If you are unsure what you have, open the file in a hex editor: a WAV starts with the ASCII characters RIFF and contains WAVE at byte 8. If you see those, rename the file to .wav and open it normally.

Why VLC and Windows Media Player refuse to play it

Media players identify files by looking for a known header or container signature. A headerless file has none, so the player either reports an unsupported format or guesses and plays noise. Some desktop tools can decode raw PCM if you supply the parameters by hand, which is useful once you have found the right settings here:

# VLC
vlc --demux=rawaud --rawaud-channels=1 --rawaud-samplerate=16000 --rawaud-fourcc=s16l file.pcm

# FFmpeg's player (mono, 16-bit signed little-endian, 16 kHz)
ffplay -f s16le -ar 16000 file.pcm

# Audacity: File → Import → Raw Data…, then pick encoding, byte order, channels and rate

The sample rate is how many samples are stored per second of audio, per channel. It determines the highest frequency the recording can contain — half the sample rate, known as the Nyquist limit. In practice the rate tells you a lot about where a file came from:

If you know the duration of a recording, the file size gives the sample rate away. Raw PCM has no compression, so the size is exactly:

bytes = sample rate × (bit depth ÷ 8) × channels × seconds

For example, a 30-second voice clip that is 960,000 bytes works out to 32,000 bytes per second. That fits 16,000 Hz × 16-bit × mono, or 8,000 Hz × 16-bit × stereo, or 32,000 Hz × 8-bit × mono. Voice clips are almost always mono, so 16 kHz / 16-bit / mono is the one to try first — and playing it settles the rest in seconds.

Bit depth is how many bits store each sample, which sets the dynamic range — roughly 6 dB per bit. But the number alone is not enough: the same 16 bits can be read as signed or unsigned, and the two bytes can be stored in either order. This player decodes the three layouts that cover the vast majority of real files:

Byte order (endianness) only matters for samples wider than 8 bits. Little-endian stores the least significant byte first; it is what x86 PCs, ARM phones and ESP32 chips produce natively, and it is what this player expects. Big-endian appears mainly in network streams — the L16 RTP payload is defined in network byte order — and in AIFF files. A byte-swapped file is easy to recognise: you hear loud hiss with the rhythm of the original audio buried inside it.

Signed vs unsigned is an 8-bit problem in practice. The WAV convention is unsigned (silence is 128), while a few systems use signed (silence is 0). Decode one as the other and every sample is shifted by half the range, which sounds like harsh distortion that gets worse, not better, in quiet passages.

If your file is in a layout the player does not read directly, convert it with SoX or FFmpeg first:

# Big-endian 16-bit → little-endian 16-bit
sox -t raw -r 16000 -e signed -b 16 -c 1 -B in.pcm -t raw -L out.pcm

# 24-bit → 16-bit (stereo, 48 kHz)
sox -t raw -r 48000 -e signed -b 24 -c 2 in.pcm -t raw -b 16 out.pcm

# 32-bit integer (e.g. ESP32 I²S dump) → 16-bit
sox -t raw -r 16000 -e signed -b 32 -c 1 in.raw -t raw -b 16 out.pcm

# G.711 µ-law → 16-bit linear
ffmpeg -f mulaw -ar 8000 -i in.ulaw -f s16le out.pcm

Most raw audio comes from a short list of places. These are the formats each one actually produces, so you can skip the guesswork:

Each wrong setting has a distinct sound, so listening carefully tells you which one to change. A wrong sample rate changes speed and pitch but leaves the audio recognisable. Wrong bit depth or byte order turns it into noise. Fix the noise first, then the speed.

Two habits make this faster. First, watch the waveform as well as listening: real speech shows distinct bursts separated by quieter gaps, while a wrong bit depth fills the display with a solid band. Second, if nothing works across every combination, the file is probably not PCM. Telephony systems often use µ-law, A-law or ADPCM, and some tools save Opus or AMR frames with a .pcm or .raw extension. Those need a decoder, not different settings.