Audio File Format Specifications

File Description: CSL NSP file
File Extension: Commonly .nsp
File Byte Order: Little-endian

Prof. Peter Kabal, MMSP Lab, ECE, McGill University: Last update: 2017-01-20
(with contributions from R. Roark)

CSL NSP File Specifications

The Computerized Speech Lab (CSL) from KayPENTAX (Kay Elemetrics) uses a proprietary file format. The NSP file format was also used for the Phonetic Datbase (PDB) from Speech Technology Research. At one time, STR had a free NSP Player: SpeakNSP.

Note that CSL files may contain both sound and bioelectric data. Typically, the first channel is sound pressure (audio) and additional channels are data such as laryngeal kinematic or aerodynamic (air pressure or air flow) obtained by instruments such as those from KayPENTAX.

CSL NSP file reading and writing is supported by the WaveSurfer package from the Speech Communication & Technology Group at KTH in Sweden; reading is supported by the AFsp package.

The NSP header format uses chunks. Each chunk consists of a chunk identifier, chunk length and chunk data.

Header Structure

The header for an NSP file starts with a pair of identifiers.

Offset	Length	Type	Contents
0	4	character	Identifier: `FORM`
4	4	character	Chunk identifier: `DS16`
8	4	unsigned integer	Chunk length
12	-	-	Chunk data

The Chunk Length is the remaining number of bytes in the chunk. For the FORM chunk, the chunk length is the file length less 12 bytes.
The chunk identifier DS16 indicates that the data are coded as 16-bit integer.
All data are in little-endian byte order.
The chunk length gives the actual length of the chunk (can be odd), but the chunk itself is zero-padded to bring it up to an even length.
All the rest of the file data is part of the DS16 chunk.
The sub-chunks are HEDR or HDR8, NOTE, SDA_, SD_B, SDAB. SD_2, ... SD_8.

`HEDR` or `HDR8` Chunk

Offset	Length	Type	Contents
0	4	character	Identifier: `HEDR` or `HDR8`
4	4	unsigned integer	Chunk length (`32` or `44`)
8	20	character	Date, e.g. `May 26 23:57:43 1995`
28	4	unsigned integer	Sampling rate
32	4	unsigned integer	Data length (bytes)
36	2	unsigned integer	Maximum absolute value for channel A; `0xFFFF` if not defined
38	2	unsigned integer	Maximum absolute value for channel B: `0xFFFF` if not defined
40	12	6 unsigned integers	`HDR8` only: Maximum absolute values for channel 3 through channel 8

`NOTE` Chunk

Offset	Length	Type	Contents
0	4	character	Identifier: `NOTE`
4	4	integer	Chunk length
8	-	character	Comment string

`SDA_`, `SD_A`, `SDAB`, `SD_2`, `SD_3`, ... , `SD_8` Chunk

Offset	Length	Type	Contents
0	4	character	Identifier: `SDA_`, `SD_B`, `SDAB`, `SD_2`, ... , `SD_8`
4	4	integer	Chunk length
8	2N	N integers	Integer data

When the chunk designator is SDAB, the two-channel data are interlaced. For all other chunk identifiers, the data values are sequential by channel. The chunk designators indicate whether left, right, both channels, or multiple channels are present.

Sample Files

addf8.nsp (47 kB): NSP file (produced by WaveSurfer), 8 kHz, 16-bit data