Wave File Format Details

Overview

Data Formats

Strings
Wave files may contain strings of text for specifying cue point labels, notes, etc. Strings are stored in a format where the first byte specifies the number of following ASCII text bytes in the string. The following bytes are of course the ASCII character bytes that make up the text string. Pascal programmers will notice that this is the same format used for Pascal strings.

File Structure

One tricky thing about RIFF file chunks is that they must be word aligned. This means that their total size must be a multiple of 2 bytes (ie. 2, 4, 6, 8, and so on). If a chunk contains an odd number of data bytes, causing it not to be word aligned, an extra padding byte with a value of zero must follow the last data byte. This extra padding byte is not counted in the chunk size, therefor a program must always word align a chunk headers size value in order to calculate the offset of the following chunk.

Wave File Header - RIFF Type Chunk

Wave File Chunks

All RIFF Chunks and therefore Wave Chunks are stored in the following format. Notice that even the above mentioned RIFF Type Chunk conforms to this format.

The rest of this document goes through the different types of Wave chunks, describing the format of their data bytes and what they mean. You can use the table of contents at the beginning of this document to help find the chunk type you are interested in.

Format Chunk - "fmt "

Chunk ID and Data Size
The chunk ID is always "fmt " (0x666D7420) and the size is the size of the standard wave format data (16 bytes) plus the size of any extra format bytes needed for the specific Wave format, if it does not contain uncompressed PCM data. Note the chunk ID string ends with the space character (0x20).

Compression Code
The first word of format data specifies the type of compression used on the Wave data included in the Wave chunk found in this "RIFF" chunk. The following is a list of the common compression codes used today.

Number of Channels
The number of channels specifies how many separate audio signals that are encoded in the wave data chunk. A value of 1 means a mono signal, a value of 2 means a stereo signal, etc.

Sample Rate
The number of sample slices per second. This value is unaffected by the number of channels.

Average Bytes Per Second
This value indicates how many bytes of wave data must be streamed to a D/A converter per second in order to play the wave file. This information is useful when determining if data can be streamed from the source fast enough to keep up with playback. This value can be easily calculated with the formula:

AvgBytesPerSec = SampleRate * BlockAlign

Block Align
The number of bytes per sample slice. This value is not affected by the number of channels and can be calculated with the formula:

BlockAlign = SignificantBitsPerSample / 8 * NumChannels

Significant Bits Per Sample
This value specifies the number of bits used to define each sample. This value is usually 8, 16, 24 or 32. If the number of bits is not byte aligned (a multiple of 8) then the number of bytes used per sample is rounded up to the nearest byte size and the unused bytes are set to 0 and ignored.

Extra Format Bytes
This value specifies how many additional format bytes follow. It does not exist if the compression code is 0 (uncompressed PCM file) but may exist and have any value for other compression types depending on what compression information is need to decode the wave data. If this value is not word aligned (a multiple of 2), padding should be added to the end of this data to word align it, but the value should remain non-aligned.

Data Chunk - "data"

Wave files usually contain only one data chunk, but they may contain more than one if they are contained within a Wave List Chunk ("wavl").

Multi-channel digital audio samples are stored as interlaced wave data which simply means that the audio samples of a multi-channel (such as stereo and surround) wave file are stored by cycling through the audio samples for each channel before advancing to the next sample time. This is done so that the audio files can be played or streamed before the entire file can be read. This is handy when playing a large file from disk (that may not completely fit into memory) or streaming a file over the Internet. The values in the diagram below would be stored in a Wave file in the order they are listed in the Value column (top to bottom).

One point about sample data that may cause some confusion is that when samples are represented with 8-bits, they are specified as unsigned values. All other sample bit-sizes are specified as signed values. For example a 16-bit sample can range from -32,768 to +32,767 with a mid-point (silence) at 0.

As mentioned earlier, all RIFF chunks (including WAVE "data" chunks) must be word aligned. If the sample data uses an odd number of bytes, a padding byte with a value of zero must be placed at the end of the sample data. The "data" chunk header's size should not include this byte.

Fact Chunk - "fact"

Format Dependant Data
There is currently only one field defined for the format dependant data. It is a single 4-byte value that specifies the number of samples in the waveform data chunk. This value can be used with the Samples Per Second value specified in the format chunk to calculate the waveforms length in seconds.

As new WAVE formats are introduced, the fact chunk will be expanded, appending fields after the defined number of samples field. Applications can use the fact chunk size to determine which fields are present in the chunk.

Wave List Chunk - "wavl"

This type of chunk is considered to be an abuse of the Wave file format by many programmers (including myself) and it's support is not recommended. Also many programs will not recognize this type of chunk, simply ignoring it. This form of compression unnecessarily complicates the structure of a Wave file and could be better implemented in other ways, including the use of several existing compression formats.

Silent Chunk - "slnt"

Number of Silent Samples
This value specifies the number of silent samples that appear in the waveform at this point in the wave list chunk.

Cue Chunk - "cue "

Chunk ID and Data Size
The chunk ID for a cue chunk is always "cue " (0x666D7420). Note that the chunk ID string ends with the space character (0x20). The chunk data size is equal to the size of the Num Cue Points value (4) plus the number of following cue points multiplied by the size of each cue point's data (24). The following formula can be used to calculate a Cue Chunk's data size:

ChunkDataSize = 4 + (NumCuePoints * 24)

Num Cue Points
This value specifies the number of following cue points in this chunk.

List of Cue Points
A list of cue points is simply a set of consecutive cue point descriptions that follow the format described below.

ID
Each cue point has a unique identification value used to associate cue points with information in other chunks. For example, a Label chunk contains text that describes a point in the wave file by referencing the associated cue point.

Position
The position specifies the sample offset associated with the cue point in terms of the sample's position in the final stream of samples generated by the play list. Said in another way, if a play list chunk is specified, the position value is equal to the sample number at which this cue point will occur during playback of the entire play list as defined by the play list's order. If no play list chunk is specified this value should be 0.

Data Chunk ID
This value specifies the four byte ID used by the chunk containing the sample that corresponds to this cue point. A Wave file with no play list is always "data". A Wave file with a play list containing both sample data and silence may be either "data" or "slnt".

Chunk Start
The Chunk Start value specifies the byte offset into the Wave List Chunk of the chunk containing the sample that corresponds to this cue point. This is the same chunk described by the Data Chunk ID value. If no Wave List Chunk exists in the Wave file, this value is 0. If a Wave List Chunk exists, this is the offset into the "wavl" chunk. The first chunk in the Wave List Chunk would be specified with a value of 0.

Block Start
The Block Start value specifies the byte offset into the "data" or "slnt" Chunk to the start of the block containing the sample. The start of a block is defined as the first byte in uncompressed PCM wave data or the last byte in compressed wave data where decompression can begin to find the value of the corresponding sample value.

Sample Offset
The Sample Offset specifies an offset into the block (specified by Block Start) for the sample that corresponds to the cue point. In uncompressed PCM waveform data, this is simply the byte offset into the "data" chunk. In compressed waveform data, this value is equal to the number of samples (may or may not be bytes) from the Block Start to the sample that corresponds to the cue point.

Playlist Chunk - "plst"

Number of Segments
This value specifies the number of following segments in the playlist chunk.

List of Segments
A list of segments is simply a set of consecutive segment descriptions that follow the format described below. The segments do not have to be in any particular order because each segments associated cue point position is used to determine the play order.

Cue Point ID
The Cue Point ID specifies the starting sample for this segment by providing the value of a Cue Point defined in the Cue Point List. The ID that associates this segment with a Cue Point must be unique to all other segment Cue Point IDs.

Length
The Segment Length specifies the number of samples to play/loop from the starting sample defined by the associated Cue Point.

Number of Repeats
The number of repeats determines how many times this segment should be looped before playback should continue onto the next segment.

Associated Data List Chunk - "list"

Type ID
The type ID is used to identify the type of associated data list and is always "adtl".

List of Text Labels and Names
The list of text labels and names is a list of assorted chunks that define text in different ways. The three main chunk types that are used in WAVE files are the Label Chunk, Note Chunk and Labeled Text Chunk.

Label Chunk - "labl"

Cue Point ID
The Cue Point ID specifies the sample point that corresponds to this text label by providing the ID of a Cue Point defined in the Cue Point List. The ID that associates this label with a Cue Point must be unique to all other label Cue Point IDs.

Text
The text is a null terminated string of characters. If the number of characters in the string is not even, padding must be appended to the string. The appended padding is not considered in the label chunk's chunk size field.

Note Chunk - "note"

Cue Point ID
The Cue Point ID specifies the sample point that corresponds to this text comment by providing the ID of a Cue Point defined in the Cue Point List. The ID that associates this label with a Cue Point must be unique to all other note chunk Cue Point IDs.

Text
The text is a null terminated string of characters. If the number of characters in the string is not even, padding must be appended to the string. The appended padding is not considered in the note chunk's chunk size field.

Labeled Text Chunk - "ltxt"

Cue Point ID
The Cue Point ID specifies the starting sample point that corresponds to this text label by providing the ID of a Cue Point defined in the Cue Point List. The ID that associates this label with a Cue Point must be unique to all other note chunk Cue Point IDs.

Sample Length
The sample length defines how many samples from the cue point the region or section spans.

Purpose ID
The purpose field specifies what the text is used for. For example a value of "scrp" means script text, and "capt" means close-caption. There are several more purpose IDs, but they are meant to be used with other types of RIFF files (not usually found in WAVE files).

Country, Language, Dialect and Code Page
These fields are used to specify information about the location and language used by the text and are typically used for queries to obtain information from the operating system.

Text
The text is a null terminated string of characters. If the number of characters in the string is not even, padding must be appended to the string. The appended padding is not considered in the note chunk's chunk size field.

Sampler Chunk - "smpl"

Manufacturer
The manufacturer field specifies the MIDI Manufacturer's Association (MMA) Manufacturer code for the sampler intended to receive this file's waveform. Each manufacturer of a MIDI product is assigned a unique ID which identifies the company. If no particular manufacturer is to be specified, a value of 0 should be used.

The value is stored with some extra information to enable translation to the value used in a MIDI System Exclusive transmission to the sampler. The high byte indicates the number of low order bytes (1 or 3) that are valid for the manufacturer code. For example, the value for Digidesign will be 0x01000013 (0x13) and the value for Microsoft will be 0x30000041 (0x00, 0x00, 0x41). See the MIDI Manufacturers List for a list.

Product
The product field specifies the MIDI model ID defined by the manufacturer corresponding to the Manufacturer field. Contact the manufacturer of the sampler to get the model ID. If no particular manufacturer's product is to be specified, a value of 0 should be used.

Sample Period
The sample period specifies the duration of time that passes during the playback of one sample in nanoseconds (normally equal to 1 / Samplers Per Second, where Samples Per Second is the value found in the format chunk).

MIDI Unity Note
The MIDI unity note value has the same meaning as the instrument chunk's MIDI Unshifted Note field which specifies the musical note at which the sample will be played at it's original sample rate (the sample rate specified in the format chunk).

MIDI Pitch Fraction
The MIDI pitch fraction specifies the fraction of a semitone up from the specified MIDI unity note field. A value of 0x80000000 means 1/2 semitone (50 cents) and a value of 0x00000000 means no fine tuning between semitones.

SMPTE Format
The SMPTE format specifies the Society of Motion Pictures and Television E time format used in the following SMPTE Offset field. If a value of 0 is set, SMPTE Offset should also be set to 0.

SMPTE Offset
The SMPTE Offset value specifies the time offset to be used for the synchronization / calibration to the first sample in the waveform. This value uses a format of 0xhhmmssff where hh is a signed value that specifies the number of hours (-23 to 23), mm is an unsigned value that specifies the number of minutes (0 to 59), ss is an unsigned value that specifies the number of seconds (0 to 59) and ff is an unsigned value that specifies the number of frames (0 to -1).

Sample Loops
The sample loops field specifies the number Sample Loop definitions in the following list. This value may be set to 0 meaning that no sample loops follow.

Sampler Data
The sampler data value specifies the number of bytes that will follow this chunk (including the entire sample loop list). This value is greater than 0 when an application needs to save additional information. This value is reflected in this chunks data size value.

List of Sample Loops
A list of sample loops is simply a set of consecutive loop descriptions that follow the format described below. The sample loops do not have to be in any particular order because each sample loop associated cue point position is used to determine the play order. The sampler chunk is optional.

Cue Point ID
The Cue Point ID specifies the unique ID that corresponds to one of the defined cue points in the cue point list. Furthermore, this ID corresponds to any labels defined in the associated data list chunk which allows text labels to be assigned to the various sample loops.

Type
The type field defines how the waveform samples will be looped.

Start
The start value specifies the byte offset into the waveform data of the first sample to be played in the loop.

End
The end value specifies the byte offset into the waveform data of the last sample to be played in the loop.

Fraction
The fractional value specifies a fraction of a sample at which to loop. This allows a loop to be fine tuned at a resolution greater than one sample. The value can range from 0x00000000 to 0xFFFFFFFF. A value of 0 means no fraction, a value of 0x80000000 means 1/2 of a sample length. 0xFFFFFFFF is the smallest fraction of a sample that can be represented.

Play Count
The play count value determines the number of times to play the loop. A value of 0 specifies an infinite sustain loop. An infinite sustain loop will continue looping until some external force interrupts playback, such as the musician releasing the key that triggered the wave's playback. All other values specify an absolute number of times to loop.

Instrument Chunk - "inst"

Unshifted Note
The unshifted note field has the same meaning as the sampler chunk's MIDI Unity Note which specifies the musical note at which the sample will be played at it's original sample rate (the sample rate specified in the format chunk).

Fine Tune
The fine tune value specifies how much the sample's pitch should be altered when the sound is played back in cents (1/100 of a semitone). A negative value means that the pitch should be played lower and a positive value means that it should be played at a higher pitch.

Gain
The gain value specifies the number of decibels to adjust the output when it is played. A value of 0dB means no change, 6dB means double the amplitude of each sample and -6dB means to halve the amplitude of each sample. Every additional +/-6dB will double or halve the amplitude again.

Low Note and High Note
The note fields specify the MIDI note range for which the waveform should be played when receiving MIDI note events (from software or triggered by a MIDI controller). This range does not need to include the Unshifted Note value.

Low Velocity and High Velocity
The velocity fields specify the range of MIDI velocities that should cause the waveform to be played. 1 being the lightest amount and 127 being the hardest.

Format Variations

• Incorrect Block Alignment value - this can be dealt with by calculating the Block Alignment with the formula mentioned above.
• Incorrect Average Samples Per Second value - this can be dealt with by calculating the Average Samples Per Second with the formula mentioned above.
• Missing word alignment padding - this can be difficult to deal with, but can be done by giving the user a warning when unrecognized chunk ID's are encountered where a one byte read offset produces a recognized chunk ID. This is not a concrete solution, but will usually work even if the program doesn't have a comprehensive list of legal IDs.