shithub: sox

Info  •  Files  •  Log  •  Branches

drop rounding in SOX_SAMPLE_TO_FLOAT_32BIT [bug #317]

The rounding of values to 24-bit integer precision serves no useful
purpose.  Drop it, allowing conversion of as many values as possible
without loss of precision.

--- a/src/sox.h

+++ b/src/sox.h

@@ -1009,14 +1009,9 @@

 Client API:

 Converts SoX native sample to a 32-bit float.

 @param d Input sample to be converted.

-@param clips Variable to increment if input sample is too large.

+@param clips The parameter is not used.

 */

-#define SOX_SAMPLE_TO_FLOAT_32BIT(d,clips) (                            \

-  LSX_USE_VAR(sox_macro_temp_double),                                   \

-  sox_macro_temp_sample = (d),                                          \

-  sox_macro_temp_sample > SOX_SAMPLE_MAX - 64 ?                         \

-    ++(clips), 1 :                                                      \

-    (((sox_macro_temp_sample + 64) & ~127) * (1.0 / (SOX_SAMPLE_MAX + 1.0))))

+#define SOX_SAMPLE_TO_FLOAT_32BIT(d,clips) ((d)*(1.0 / (SOX_SAMPLE_MAX + 1.0)))

 /**

 Client API: