fileio.c File Reference

Audio file I/O and feature vector writing. More...

#include "fileio.h"

Include dependency graph for fileio.c:

Go to the source code of this file.

Data Structures
struct	FIO_HTKheader
	HTK file header structure (12 bytes total). More...

Macros
#define	SWAP(x)   swap_bytes(&(x), sizeof(x))
	Convenience macro: swap the bytes of variable x.

Functions
static int	is_bigendian (void)
	Check if the host is big-endian.
static void	swap_bytes (void *pv, size_t n)
	Reverse the bytes of a value in-place (e.g.
int	FIO_read_audio (const char *infile, float **indata, size_t *datalen, unsigned *samprate, unsigned donorm)
	Read a single-channel audio file into a float array.
int	FIO_write_npy (const char *outfile, const float **outvecs, size_t nvecs, size_t veclen)
	Write a 2D float32 array in NumPy .npy v1.0 format.
int	FIO_write_safetensors (const char *outfile, const float **outvecs, size_t nvecs, size_t veclen)
	Write a 2D float32 array in safetensors format.
int	FIO_write_wav (const char *outfile, const float *data, size_t datalen, unsigned samprate)
	Write mono float audio to a WAV file.
int	FIO_write_htk_feats (const char *outfile, const float **outvecs, size_t nvecs, size_t veclen, unsigned vecsamprate)
	Write feature vectors in HTK binary file format.

Detailed Description

Audio file I/O and feature vector writing.

This module reads audio using libsndfile (which supports dozens of formats including WAV, FLAC, AIFF, and Ogg), writes audio to WAV, and writes feature vectors in NumPy .npy, safetensors, and HTK formats.

Definition in file fileio.c.

Macro Definition Documentation

SWAP

#define SWAP

(

x

)

swap_bytes(&(x), sizeof(x))

Convenience macro: swap the bytes of variable x.

Definition at line 63 of file fileio.c.

Function Documentation

FIO_read_audio()

int FIO_read_audio	(	const char *	infile,
		float **	indata,
		size_t *	datalen,
		unsigned *	samprate,
		unsigned	donorm
	)

Read a single-channel audio file into a float array.

Read a single-channel audio file into memory.

Supported formats include everything that libsndfile can open: WAV, FLAC, AIFF, OGG, and many more.

Parameters

infile	Path to the audio file.
indata	Output: pointer to allocated float array with samples. The caller is responsible for calling free() on this.
datalen	Output: total number of samples read.
samprate	Output: sample rate in Hz.
donorm	1 to normalise samples to [-1.0, 1.0]; 0 for raw values.

Returns

0 on success, -1 on error.

Definition at line 83 of file fileio.c.

FIO_write_htk_feats()

int FIO_write_htk_feats	(	const char *	outfile,
		const float **	outvecs,
		size_t	nvecs,
		size_t	veclen,
		unsigned	vecsamprate
	)

Write feature vectors in HTK binary file format.

Write feature vectors in HTK binary format.

The HTK format is: [12-byte header][vector 1][vector 2]...[vector N]

Each vector is a sequence of big-endian 32-bit floats.

Parameters

outfile	Output file path.
outvecs	Array of nvecs pointers, each to veclen floats.
nvecs	Number of feature vectors.
veclen	Number of float elements per vector.
vecsamprate	Feature sampling rate in Hz.

Returns

0 on success, -1 on error.

Definition at line 323 of file fileio.c.

FIO_write_npy()

int FIO_write_npy	(	const char *	outfile,
		const float **	outvecs,
		size_t	nvecs,
		size_t	veclen
	)

Write a 2D float32 array in NumPy .npy v1.0 format.

Produces a file readable by numpy.load(). Data is stored as little-endian float32, row-major (C order).

Parameters

outfile	Path to the output .npy file.
outvecs	Array of nvecs pointers, each pointing to veclen floats.
nvecs	Number of feature vectors (rows).
veclen	Number of floats per vector (columns).

Returns

0 on success, -1 on error.

Definition at line 167 of file fileio.c.

FIO_write_safetensors()

int FIO_write_safetensors	(	const char *	outfile,
		const float **	outvecs,
		size_t	nvecs,
		size_t	veclen
	)

Write a 2D float32 array in safetensors format.

Produces a file readable by the safetensors Python library. The tensor is stored under the key "features" as little-endian float32.

Parameters

outfile	Path to the output .safetensors file.
outvecs	Array of nvecs pointers, each pointing to veclen floats.
nvecs	Number of feature vectors (rows).
veclen	Number of floats per vector (columns).

Returns

0 on success, -1 on error.

Definition at line 236 of file fileio.c.

FIO_write_wav()

int FIO_write_wav	(	const char *	outfile,
		const float *	data,
		size_t	datalen,
		unsigned	samprate
	)

Write mono float audio to a WAV file.

Uses IEEE float format (SF_FORMAT_FLOAT) for lossless DSP round-trips.

Parameters

outfile	Path to the output .wav file.
data	Audio samples.
datalen	Number of samples.
samprate	Sampling rate in Hz.

Returns

0 on success, -1 on error.

Definition at line 282 of file fileio.c.

is_bigendian()

static int is_bigendian ( void  )

static

Check if the host is big-endian.

Definition at line 45 of file fileio.c.

swap_bytes()

static void swap_bytes ( void *  pv, size_t  n  )

static

Reverse the bytes of a value in-place (e.g.

convert little-endian to big).

Definition at line 52 of file fileio.c.