The libsndfile API

Libsndfile is a library designed to allow the reading and writing of many different sampled sound file formats (such as MS Windows WAV and the Apple/SGI AIFF format) through one standard library interface.

During read and write operations, formats are seamlessly converted between the format the application program has requested or supplied and the file's data format. The application programmer can remain blissfully unaware of issues such as file endian-ness and data format. See Note 1 and Note 2.

Every effort is made to keep these documents up-to-date, error free and unambiguous. However, since maintaining the documentation is the least fun part of working on libsndfile, these docs can and do fall behind the behaviour of the library. If any errors, omissions or ambiguities are found, please notify me (erikd) at mega-nerd dot com.

To supplement this reference documentation, there are simple example programs included in the source code tarball. The test suite which is also part of the source code tarball is also a good place to look for the correct usage of the library functions.

Finally, if you think there is some feature missing from libsndfile, check that it isn't already implemented (and documented) here.

Synopsis

File Open Function

The sf_open() function opens the sound file at the specified path. The filename is byte encoded, but may be utf-8 on Linux, while on Mac OS X it will use the filesystem character set. On Windows, there is also a Windows specific sf_wchar_open() that takes a UTF16_BE encoded filename.

The SF_INFO structure is for passing data between the calling function and the library when opening a file for reading or writing. It is defined in sndfile.h as follows:

The mode parameter for this function can be any one of the following three values:

When opening a file for read, the format field should be set to zero before calling sf_open(). The only exception to this is the case of RAW files where the caller has to set the samplerate, channels and format fields to valid values. All other fields of the structure are filled in by the library.

When opening a file for write, the caller must fill in structure members samplerate, channels, and format.

The format field in the above SF_INFO structure is made up of the bit-wise OR of a major format type (values between 0x10000 and 0x08000000), a minor format type (with values less than 0x10000) and an optional endian-ness value. The currently understood formats are listed in sndfile.h as follows and also include bitmasks for separating major and minor file types. Not all combinations of endian-ness and major and minor file types are valid.

Every call to sf_open() should be matched with a call to sf_close() to free up memory allocated during the call to sf_open().

On success, the sf_open function returns a non-NULL pointer which should be passed as the first parameter to all subsequent libsndfile calls dealing with that audio file. On fail, the sf_open function returns a NULL pointer. An explanation of the error can obtained by passing NULL to sf_strerror.

File Descriptor Open

Note: On Microsoft Windows, this function does not work if the application and the libsndfile DLL are linked to different versions of the Microsoft C runtime DLL.

The second open function takes a file descriptor of a file that has already been opened. Care should be taken to ensure that the mode of the file represented by the descriptor matches the mode argument. This function is useful in the following circumstances:

Every call to sf_open_fd() should be matched with a call to sf_close() to free up memory allocated during the call to sf_open_fd().

When sf_close() is called, the file descriptor is only closed if the close_desc parameter was TRUE when the sf_open_fd() function was called.

On success, the sf_open_fd function returns a non-NULL pointer which should be passed as the first parameter to all subsequent libsndfile calls dealing with that audio file. On fail, the sf_open_fd function returns a NULL pointer.

Virtual File Open Function

Opens a soundfile from a virtual file I/O context which is provided by the caller. This is usually used to interface libsndfile to a stream or buffer based system. Apart from the sfvirtual and the user_data parameters this function behaves like sf_open.

Libsndfile calls the callbacks provided by the SF_VIRTUAL_IO structure when opening, reading and writing to the virtual file context. The user_data pointer is a user defined context which will be available in the callbacks.

sf_vio_get_filelen

sf_vio_seek

The virtual file context must seek to offset using the seek mode provided by whence which is one of

sf_vio_read

The virtual file context must copy ("read") "count" bytes into the buffer provided by ptr and return the count of actually copied bytes.

sf_vio_write

The virtual file context must process "count" bytes stored in the buffer passed with ptr and return the count of actually processed bytes.

sf_vio_tell

Format Check Function

This function allows the caller to check if a set of parameters in the SF_INFO struct is valid before calling sf_open (SFM_WRITE).

File Seek Functions

The file seek functions work much like lseek in unistd.h with the exception that the non-audio data is ignored and the seek only moves within the audio data section of the file. In addition, seeks are defined in number of (multichannel) frames. Therefore, a seek in a stereo file from the current position forward with an offset of 1 would skip forward by one sample of both channels.

like lseek(), the whence parameter can be any one of the following three values:

Internally, libsndfile keeps track of the read and write locations using separate read and write pointers. If a file has been opened with a mode of SFM_RDWR, bitwise OR-ing the standard whence values above with either SFM_READ or SFM_WRITE allows the read and write pointers to be modified separately. If the SEEK_* values are used on their own, the read and write pointers are both modified.

Note that the frames offset can be negative and in fact should be when SEEK_END is used for the whence parameter.

sf_seek will return the offset in (multichannel) frames from the start of the audio data or -1 if an error occured (ie an attempt is made to seek beyond the start or end of the file).

Error Reporting Functions

This function returns the current error number for the given SNDFILE. The error number may be one of the following:

or any one of many other internal error values. Applications should only test the return value against error values defined in <sndfile.h> as the internal error values are subject to change at any time. For errors not in the above list, the function sf_error_number() can be used to convert it to an error string.

The error functions sf_strerror() and sf_error_number() convert the library's internal error enumerations into text strings.

The functions sf_perror() and sf_error_str() are deprecated and will be dropped from the library at some later date.

File Close Function

The close function closes the file, deallocates its internal buffers and returns 0 on success or an error value otherwise.

Write Sync Function

If the file is opened SFM_WRITE or SFM_RDWR, call the operating system's function to force the writing of all file cache buffers to disk. If the file is opened SFM_READ no action is taken.

File Read Functions

The file read functions fill the array pointed to by ptr with the requested number of items or frames.

For the frames-count functions, the frames parameter specifies the number of frames. A frame is just a block of samples, one for each channel. Care must be taken to ensure that there is enough space in the array pointed to by ptr, to take (frames * channels) number of items (shorts, ints, floats or doubles).

For the items-count functions, the items parameter must be an integer product of the number of channels or an error will occur. Here, an item is just a sample.

Note: The only difference between the "items" and "frames" versions of each read function is the units in which the object count is specified - calling sf_readf_short with a count argument of N, on a SNDFILE with C channels, is the same as calling sf_read_short with a count argument of N*C. The buffer pointed to by "ptr" should be the same number of bytes in each case.

Note: The data type used by the calling program and the data format of the file do not need to be the same. For instance, it is possible to open a 16 bit PCM encoded WAV file and read the data using sf_read_float(). The library seamlessly converts between the two formats on-the-fly. See Note 1.

The sf_read_XXXX and sf_readf_XXXX functions return the number of items or frames read, respectively. Unless the end of the file was reached during the read, the return value should equal the number of objects requested. Attempts to read beyond the end of the file will not result in an error but will cause the read functions to return less than the number of objects requested or 0 if already at the end of the file. When the buffer is not is not completely filled, unused buffer space is filled by zeroes.

File Write Functions

The file write functions write the data in the array pointed to by ptr to the file.

For items-count functions, the items parameter specifies the size of the array and must be an integer product of the number of channels or an error will occur.

For the frames-count functions, the array is expected to be large enough to hold a number of items equal to the product of frames and the number of channels.

As with the read functions above, the only difference in the items and frames version of each write function is the units in which the buffer size is specified. Again, the data type used by the calling program and the data format of the file do not need to be the same (Note 1).

The sf_write_XXXX and sf_writef_XXXX functions respectively return the number of items or frames written (which should be the same as the items or frames parameter).

Raw File Read and Write Functions

Note: Unless you are writing an external decoder/encode that uses libsndfile to handle the file headers, you should not be using these functions.

The raw read and write functions read raw audio data from the audio file (not to be confused with reading RAW header-less PCM files). The number of bytes read or written must always be an integer multiple of the number of channels multiplied by the number of bytes required to represent one sample from one channel.

The raw read and write functions return the number of bytes read or written (which should be the same as the bytes parameter).

Note : The result of using of both regular reads/writes and raw reads/writes on compressed file formats other than SF_FORMAT_ALAW and SF_FORMAT_ULAW is undefined.

Functions for Reading and Writing String Data

These functions allow strings to be set on files opened for write and to be retrieved from files opened for read where supported by the given file type. The str_type parameter can be any one of the following string types:

The sf_get_string() function returns the specified string if it exists and a NULL pointer otherwise. In addition to the string ids above, SF_STR_FIRST (== SF_STR_TITLE) and SF_STR_LAST (always the same as the highest numbers string id) are also available to allow iteration over all the available string ids.

The sf_set_string() function sets the string data. It returns zero on success and non-zero on error. The error code can be converted to a string using sf_error_number().

Strings passed to and retrieved from these two functions are assumed to be utf-8. However, while formats like Ogg/Vorbis and FLAC fully support utf-8, others like WAV and AIFF officially only support ASCII. Writing utf-8 strings to WAV and AIF files with libsndfile will work when read back with libsndfile, but may not work with other programs.

The suggested method of dealing with tags retrived using sf_get_string() is to assume they are utf-8. Similarly if you have a string in some exotic format like utf-16, it should be encoded to utf-8 before being written using libsndfile.

Functions to get and set chunks from within a sound file

These functions allow the getting and setting of chunks within a sound file (for those formats which allow it).

These functions fail safely. Specifically, they will not allow you to overwrite existing chunks or add extra versions of format specific reserved chunks but should allow you to retrieve any and all chunks (may not be implemented for all chunks or all file formats).

sf_set_chunk

      int sf_set_chunk (SNDFILE * sndfile, const SF_CHUNK_INFO * chunk_info) ;

Set the specified chunk info (must be done before any audio data is written to the file). This will fail for format specific reserved chunks. The chunk_info->data pointer must be valid until the file is closed.

The SF_CHUNK_INFO struct is documented as follows:

      struct SF_CHUNK_INFO
      {     char        id [64] ;   /* The chunk identifier. */
            unsigned    id_size ;   /* The size of the chunk identifier. */
            unsigned    datalen ;   /* The size of that data. */
            void        *data ;     /* Pointer to the data. */
      } ;
      
      typedef struct SF_CHUNK_INFO SF_CHUNK_INFO ;

sf_set_chunk returns SF_ERR_NO_ERROR on success or non-zero on failure.

sf_get_chunk_iterator

      SF_CHUNK_ITERATOR *
      sf_get_chunk_iterator (SNDFILE * sndfile, const SF_CHUNK_INFO * chunk_info) ;

Get an iterator for all chunks matching chunk_info.

SF_CHUNK_ITERATOR is an opaque structure to an iterator over the all chunks of a given id and defined as follows:

typedef	struct SF_CHUNK_ITERATOR SF_CHUNK_ITERATOR ;

The iterator will point to the first chunk matching chunk_info. Chunks are matching, if (chunk_info->id) matches the first ( chunk_info->id_size) bytes of a chunk found in the SNDFILE* handle. If chunk_info is NULL, an iterator to all chunks in the SNDFILE* handle is returned. The values of chunk_info->datalen and chunk_info->data are ignored. If no matching chunks are found in the sndfile, NULL is returned.

The returned iterator will stay valid until one of the following occurs:

The sndfile is closed.
A new chunk is added using sf_set_chunk() .
Another chunk iterator function is called on the same SNDFILE* handle that causes the iterator to be modified.

The memory for the iterator belongs to the SNDFILE* handle and is freed when sf_close() is called.

sf_next_chunk_iterator

      sf_next_chunk_iterator (SF_CHUNK_ITERATOR * iterator) ;

Iterate through chunks by incrementing the iterator.

Increments the iterator and returns a handle to the new one. After this call, iterator will no longer be valid, and you must use the newly returned handle from now on. The returned handle can be used to access the next chunk matching the criteria as defined in sf_get_chunk_iterator(). If iterator points to the last chunk, this will free all resources associated with iterator and return NULL. The returned iterator will stay valid until sf_get_next_chunk_iterator is called again, the sndfile is closed or a new chunk us added.

sf_get_chunk_size

      int
      sf_get_chunk_size (const SF_CHUNK_ITERATOR * it, SF_CHUNK_INFO * chunk_info) ;

Get the size of the specified chunk.

If the specified chunk exists, the size will be returned in the datalen field of the SF_CHUNK_INFO struct. Additionally, the id of the chunk will be copied to the id field of the SF_CHUNK_INFO struct and it's id_size field will be updated accordingly.

If the chunk doesn't exist chunk_info->datalen will be zero, and the id and id_size fields will be undefined.

The function will return SF_ERR_NO_ERROR on success or non-zero on failure.

sf_get_chunk_data

      int
      sf_get_chunk_data (const SF_CHUNK_ITERATOR * it, SF_CHUNK_INFO * chunk_info) ;

Get the specified chunk data.

If the specified chunk exists, up to chunk_info->datalen bytes of the chunk data will be copied into the chunk_info->data buffer (allocated by the caller) and the chunk_info->datalen field updated to reflect the size of the data. The id and id_size field will be updated according to the retrieved chunk. If the chunk doesn't exist chunk_info->datalen will be zero, and the id and id_size fields will be undefined.

The function will return SF_ERR_NO_ERROR on success or non-zero on failure.

Note 1

When converting between integer PCM formats of differing size (e.g. using sf_read_int() to read a 16 bit PCM encoded WAV file) libsndfile obeys one simple rule:

Whenever integer data is moved from one sized container to another sized container, the most significant bit in the source container will become the most significant bit in the destination container.

When converting between integer data and floating point data, different rules apply. The default behaviour when reading floating point data (sf_read_float() or sf_read_double ()) from a file with integer data is normalisation. Regardless of whether data in the file is 8, 16, 24 or 32 bit wide, the data will be read as floating point data in the range [-1.0, 1.0]. Similarly, data in the range [-1.0, 1.0] will be written to an integer PCM file so that a data value of 1.0 will be the largest allowable integer for the given bit width. This normalisation can be turned on or off using the sf_command interface.

Note 2

Reading a file containg floating point data (allowable with WAV, AIFF, AU and other file formats) using integer read methods (sf_read_short() or sf_read_int()) can produce unexpected results. For instance the data in the file may have a maximum absolute value < 1.0 which would mean that all sample values read from the file will be zero. In order to read these files correctly using integer read methods, it is recommended that you use the sf_command interface, a command of SFC_SET_SCALE_FLOAT_INT_READ and a parameter of SF_TRUE to force correct scaling.

libsndfile

Synopsis

File Open Function

File Descriptor Open

Virtual File Open Function

sf_vio_get_filelen

sf_vio_seek

sf_vio_read

sf_vio_write

sf_vio_tell

Format Check Function

File Seek Functions

Error Reporting Functions

File Close Function

Write Sync Function

File Read Functions

File Write Functions

Raw File Read and Write Functions

Functions for Reading and Writing String Data

Function for retrieving library version

Function for retrieving current byterate