Encoder testing

This commit is contained in:
2026-06-08 12:14:32 +02:00
parent 444d62edac
commit d6aa880104
6 changed files with 261 additions and 61 deletions
+9 -9
View File
@@ -119,17 +119,17 @@
(ensure-open! fmt)
(set! frames-written (+ frames-written ((audio-encoder-write encoder) backend-handle fmt buffer buf-len))))
(define (ensure-flac-converter! input-format)
;; FLAC encoding may be used as a sample-rate conversion target, for example
;; 96 kHz -> 48 kHz. That conversion is not a property of libFLAC itself;
;; it must happen on the decoded PCM stream before process_interleaved.
(when (and (eq? kind 'flac)
(eq? converter #f)
(pcm-conversion-needed? input-format settings))
(set! converter (make-pcm-converter input-format settings))))
(define (ensure-converter! input-format)
;; FLAC may need conversion because the caller requested a target sample
;; rate or bit depth. Opus is deliberately not routed through this
;; converter by default: libopusenc accepts the source input rate and has
;; its own resampler, and opus-encoder.rkt feeds it float PCM directly.
(when (and (eq? kind 'flac) (eq? converter #f))
(when (pcm-conversion-needed? input-format settings)
(set! converter (make-pcm-converter input-format settings)))))
(define (write-converted! input-format buffer buf-len)
(ensure-flac-converter! input-format)
(ensure-converter! input-format)
(cond [converter
(let-values (((out out-samples) (pcm-converter-convert converter buffer buf-len input-format)))
(when (> out-samples 0)
+158
View File
@@ -0,0 +1,158 @@
#lang racket/base
(require "audio-encoder.rkt"
"tests.rkt"
simple-log
racket/cmdline
racket/file
racket/path
racket/string)
(provide encoder-test
encoder-test-opus
encoder-test-flac)
(define (setting-value v)
(cond ((or (eq? v #f) (eq? v 'source)) 'source)
((string? v)
(let ((s (string-downcase v)))
(if (string=? s "source")
'source
(let ((n (string->number v)))
(if n n (raise-argument-error 'encoder-test "number or source" v))))))
(else v)))
(define (encoder-symbol v)
(cond ((symbol? v) v)
((string? v) (string->symbol (string-downcase v)))
(else (raise-argument-error 'encoder-test "encoder name" v))))
(define (default-output-file encoder)
(build-path (find-system-path 'temp-dir)
(format "racket-audio-encoder-test.~a"
(case encoder
((opus) "opus")
((flac) "flac")
(else (raise-argument-error 'encoder-test "opus or flac" encoder))))))
(define (opus-settings bitrate-kbps sample-rate)
(if (eq? sample-rate 'source)
(hash 'bitrate (* bitrate-kbps 1000)
'vbr? #t
'complexity 10)
(hash 'bitrate (* bitrate-kbps 1000)
'vbr? #t
'complexity 10
'sample-rate sample-rate)))
(define (flac-settings compression-level sample-rate bits-per-sample)
(let ((h (make-hash)))
(hash-set! h 'compression-level compression-level)
(unless (eq? sample-rate 'source) (hash-set! h 'sample-rate sample-rate))
(unless (eq? bits-per-sample 'source) (hash-set! h 'bits-per-sample bits-per-sample))
h))
(define (format-summary fmt)
(if (hash? fmt)
(format "rate=~a, channels=~a, bits=~a, frames=~a"
(hash-ref fmt 'sample-rate "?")
(hash-ref fmt 'channels "?")
(hash-ref fmt 'bits-per-sample "?")
(hash-ref fmt 'total-frames "?"))
"unknown"))
(define (display-result result)
(displayln "")
(displayln "Encoder result")
(displayln "--------------")
(displayln (format "encoder : ~a" (hash-ref result 'encoder '?)))
(displayln (format "input : ~a" (hash-ref result 'input '?)))
(displayln (format "output : ~a" (hash-ref result 'output '?)))
(displayln (format "frames written : ~a" (hash-ref result 'frames-written '?)))
(displayln (format "input format : ~a" (format-summary (hash-ref result 'input-format #f))))
(displayln (format "output format : ~a" (format-summary (hash-ref result 'output-format #f))))
result)
(define (encoder-test input-file output-file encoder settings #:copy-tags? [copy-tags? #t])
(let* ((enc (encoder-symbol encoder))
(out (if output-file output-file (default-output-file enc))))
(when (file-exists? out) (delete-file out))
(displayln (format "Encoding ~a" input-file))
(displayln (format " -> ~a" out))
(displayln (format "encoder : ~a" enc))
(displayln (format "settings: ~a" settings))
(display-result (audio-encode input-file out settings #:encoder enc #:copy-tags? copy-tags?))))
(define (encoder-test-opus [input-file test-file3]
[output-file #f]
#:bitrate-kbps [bitrate-kbps 160]
#:sample-rate [sample-rate 'source]
#:copy-tags? [copy-tags? #t])
(encoder-test input-file output-file 'opus
(opus-settings bitrate-kbps (setting-value sample-rate))
#:copy-tags? copy-tags?))
(define (encoder-test-flac [input-file test-file3]
[output-file #f]
#:compression-level [compression-level 8]
#:sample-rate [sample-rate 'source]
#:bits-per-sample [bits-per-sample 'source]
#:copy-tags? [copy-tags? #t])
(encoder-test input-file output-file 'flac
(flac-settings compression-level
(setting-value sample-rate)
(setting-value bits-per-sample))
#:copy-tags? copy-tags?))
(module+ main
(sl-log-to-display)
(define encoder 'opus)
(define input-file test-file3)
(define output-file #f)
(define copy-tags? #t)
(define bitrate-kbps 160)
(define compression-level 8)
(define sample-rate 'source)
(define bits-per-sample 'source)
(command-line
#:program "encoder-test.rkt"
#:once-each
(("-e" "--encoder") e "Encoder: opus or flac. Default: opus."
(set! encoder (encoder-symbol e)))
(("-i" "--input") f "Input audio file. Default: tests.rkt test-file3."
(set! input-file f))
(("-o" "--output") f "Output audio file. Default: temp test file."
(set! output-file f))
(("--sample-rate") r "Target sample rate, e.g. 48000, or source. Default: source."
(set! sample-rate (setting-value r)))
(("--bits-per-sample") b "Target FLAC bits per sample, e.g. 16/24, or source. Default: source."
(set! bits-per-sample (setting-value b)))
(("--bitrate-kbps") b "Opus bitrate in kbps. Default: 160."
(set! bitrate-kbps (or (string->number b)
(raise-argument-error 'encoder-test "number" b))))
(("--compression-level") n "FLAC compression level. Default: 8."
(set! compression-level (or (string->number n)
(raise-argument-error 'encoder-test "number" n))))
(("--no-tags") "Do not copy tags/pictures to the output file."
(set! copy-tags? #f))
#:args rest
(cond ((null? rest) (void))
((null? (cdr rest)) (set! input-file (car rest)))
((null? (cddr rest)) (set! input-file (car rest)) (set! output-file (cadr rest)))
(else (raise-user-error 'encoder-test "too many positional arguments: ~a" rest))))
(case encoder
((opus)
(encoder-test-opus input-file output-file
#:bitrate-kbps bitrate-kbps
#:sample-rate sample-rate
#:copy-tags? copy-tags?))
((flac)
(encoder-test-flac input-file output-file
#:compression-level compression-level
#:sample-rate sample-rate
#:bits-per-sample bits-per-sample
#:copy-tags? copy-tags?))
(else (raise-argument-error 'encoder-test "opus or flac" encoder))))
+15 -7
View File
@@ -31,6 +31,9 @@
(verify? . #f)
(blocksize . 0))))
(define (source-value v source)
(if (eq? v 'source) source v))
(define (safe-flac-bits bits)
(cond [(and (integer? bits) (or (= bits 8) (= bits 12) (= bits 16) (= bits 20) (= bits 24))) bits]
[(and (integer? bits) (< bits 16)) 16]
@@ -41,13 +44,18 @@
(h (hash-merge base settings))
;; In encoder settings, 'sample-rate means the requested output rate.
;; 'target-sample-rate is accepted as an explicit alias for readability.
(rate (hash-ref/default h 'target-sample-rate
(hash-ref/default h 'sample-rate (hash-ref format 'sample-rate))))
(channels (hash-ref/default h 'target-channels
(hash-ref/default h 'channels (hash-ref format 'channels))))
(bits0 (hash-ref/default h 'target-bits-per-sample
(hash-ref/default h 'bits-per-sample
(hash-ref/default format 'bits-per-sample 24))))
(source-rate (hash-ref format 'sample-rate))
(source-channels (hash-ref format 'channels))
(source-bits (hash-ref/default format 'bits-per-sample 24))
(rate (source-value (hash-ref/default h 'target-sample-rate
(hash-ref/default h 'sample-rate source-rate))
source-rate))
(channels (source-value (hash-ref/default h 'target-channels
(hash-ref/default h 'channels source-channels))
source-channels))
(bits0 (source-value (hash-ref/default h 'target-bits-per-sample
(hash-ref/default h 'bits-per-sample source-bits))
source-bits))
(bits (safe-flac-bits bits0))
(total (hash-ref/default h 'total-samples (hash-ref/default format 'total-samples #f))))
(hash-set! h 'sample-rate rate)
+1 -1
View File
@@ -703,7 +703,7 @@
(define (bool->flac-bool v) (if v 1 0))
(define (native-signed-ref bs start bytes)
(integer-bytes->integer bs #t (system-big-endian?) start (+ start bytes)))
(int-bytes->integer bs #t (system-big-endian?) start (+ start bytes)))
(define (scale-sample sample in-bits out-bits)
(cond [(> in-bits out-bits) (arithmetic-shift sample (- out-bits in-bits))]
+55 -31
View File
@@ -11,11 +11,28 @@
opus-encoder-finish)
;; libopusenc handles the Ogg container, OpusHead and OpusTags. The Racket
;; side only feeds interleaved signed 16-bit PCM to ope_encoder_write().
;; side feeds interleaved floating-point PCM to ope_encoder_write_float().
;; The input rate passed to ope_encoder_create_file is the source PCM rate;
;; libopusenc performs the required Opus resampling internally.
;; Load libogg and libopus explicitly before libopusenc. This matters on
;; Windows, where libopusenc.dll may not reliably find its dependent DLLs
;; unless they have already been resolved through the same search path.
(define libogg
(get-lib (case (system-type 'os)
[(windows) '("ogg")]
[else '("ogg" "libogg")])
'(#f)))
(define libopus
(get-lib (case (system-type 'os)
[(windows) '("opus")]
[else '("opus" "libopus")])
'(#f)))
(define libopusenc
(get-lib (case (system-type 'os)
[(windows) '("opusenc")]
[(windows) '("libopusenc")]
[else '("opusenc" "libopusenc")])
'(#f)))
@@ -37,7 +54,7 @@
(_fun _string/utf-8 _OggOpusComments _int32 _int _int (err : (_ptr o _int))
-> (enc : _OggOpusEnc)
-> (values enc err))))
(define ope_encoder_write (ffi-proc "ope_encoder_write" (_fun _OggOpusEnc _bytes _int -> _int)))
(define ope_encoder_write_float (ffi-proc "ope_encoder_write_float" (_fun _OggOpusEnc _pointer _int -> _int)))
(define ope_encoder_drain (ffi-proc "ope_encoder_drain" (_fun _OggOpusEnc -> _int)))
(define ope_encoder_destroy (ffi-proc "ope_encoder_destroy" (_fun _OggOpusEnc -> _void)))
(define ope_strerror (ffi-proc "ope_strerror" (_fun _int -> _string/utf-8)))
@@ -56,8 +73,8 @@
(define OPUS_SIGNAL_MUSIC 3002)
(define (opus-encoder-available?)
(and libopusenc ope_comments_create ope_comments_destroy ope_encoder_create_file
ope_encoder_write ope_encoder_drain ope_encoder_destroy ope_strerror #t))
(and libogg libopus libopusenc ope_comments_create ope_comments_destroy ope_encoder_create_file
ope_encoder_write_float ope_encoder_drain ope_encoder_destroy ope_strerror #t))
(define-struct opus-encoder-handle (enc comments settings format file) #:transparent)
@@ -89,21 +106,24 @@
(complexity . 10)
(comment-padding . 512))))
(define (valid-opus-rate? rate)
(or (= rate 8000) (= rate 12000) (= rate 16000) (= rate 24000) (= rate 48000)))
(define (signal->int v)
(cond [(or (eq? v 'auto) (eq? v #f)) OPUS_AUTO]
[(eq? v 'voice) OPUS_SIGNAL_VOICE]
[(eq? v 'music) OPUS_SIGNAL_MUSIC]
[else (raise-argument-error 'opus-signal "(or/c 'auto 'voice 'music)" v)]))
(define (source-value v source)
(if (eq? v 'source) source v))
(define (opus-encoder-prepare-settings settings format)
(let* ((h (hash-merge (opus-encoder-default-settings) settings))
(rate (hash-ref/default h 'sample-rate (hash-ref format 'sample-rate)))
(channels (hash-ref/default h 'channels (hash-ref format 'channels))))
(unless (valid-opus-rate? rate)
(error 'opus-encoder-open "Opus input sample rate must be 8000, 12000, 16000, 24000 or 48000 Hz; got ~a. Resample before calling libopusenc." rate))
(rate (source-value (hash-ref/default h 'sample-rate (hash-ref format 'sample-rate))
(hash-ref format 'sample-rate)))
(channels (source-value (hash-ref/default h 'channels (hash-ref format 'channels))
(hash-ref format 'channels))))
;; Do not apply the low-level libopus sample-rate restriction here.
;; libopusenc accepts the input rate and performs the required resampling
;; internally; 44100 Hz input is therefore valid.
(when (> channels 2)
(error 'opus-encoder-open "this first direct libopusenc backend only supports mono/stereo input; got ~a channels" channels))
(hash-set! h 'sample-rate rate)
@@ -137,7 +157,8 @@
(hash-keys ch))))))
(define (opus-encoder-open output-file settings format)
(unless (opus-encoder-available?) (error 'opus-encoder-open "libopusenc could not be loaded"))
(unless (opus-encoder-available?)
(error 'opus-encoder-open "libopusenc or one of its dependent libraries (ogg/opus) could not be loaded"))
(let* ((file (if (path? output-file) (path->string output-file) output-file))
(resolved (opus-encoder-prepare-settings settings format))
(comments (ope_comments_create)))
@@ -151,35 +172,38 @@
(make-opus-encoder-handle enc comments resolved format file))))
(define (native-signed-ref bs start bytes)
(integer-bytes->integer bs #t (system-big-endian?) start (+ start bytes)))
;; Racket's integer-bytes->integer only supports 1, 2, 4 and 8 bytes.
;; The FLAC decoder legitimately produces 24-bit PCM as three bytes per
;; sample, so use the package helper that handles that case.
(int-bytes->integer bs #t (system-big-endian?) start (+ start bytes)))
(define (sample->s16 sample in-bits)
(cond [(> in-bits 16) (arithmetic-shift sample (- 16 in-bits))]
[(< in-bits 16) (arithmetic-shift sample (- 16 in-bits))]
[else sample]))
(define (sample->float sample in-bits)
(let* ((scale (expt 2 (sub1 in-bits)))
(v (/ sample scale)))
(cond [(< v -1.0) -1.0]
[(> v 1.0) 1.0]
[else (exact->inexact v)])))
(define (write-s16-native! out offset sample)
(integer->integer-bytes sample 2 #t (system-big-endian?) out offset))
(define (pcm-bytes->s16 buffer size in-bits)
(define (pcm-bytes->float-pointer buffer size in-bits)
(let* ((in-bytes (quotient in-bits 8))
(sample-count (quotient size in-bytes))
(out (make-bytes (* sample-count 2))))
(ptr (malloc _float sample-count 'atomic-interior)))
(for ([i (in-range sample-count)])
(let* ((in-off (* i in-bytes))
(out-off (* i 2))
(sample (native-signed-ref buffer in-off in-bytes)))
(write-s16-native! out out-off (sample->s16 sample in-bits))))
out))
(ptr-set! ptr _float i (sample->float sample in-bits))))
(values ptr sample-count)))
(define (opus-encoder-write handle buf-info buffer buf-len)
(let* ((settings (opus-encoder-handle-settings handle))
(channels (hash-ref settings 'channels))
(in-bits (hash-ref/default buf-info 'bits-per-sample 16))
(pcm (if (= in-bits 16) buffer (pcm-bytes->s16 buffer buf-len in-bits)))
(frames (quotient (quotient (bytes-length pcm) 2) channels)))
(check-ope 'opus-encoder-write (ope_encoder_write (opus-encoder-handle-enc handle) pcm frames))
frames))
(in-bits (hash-ref/default buf-info 'pcm-bits-per-sample
(hash-ref/default buf-info 'bits-per-sample 16))))
(let-values (((pcm sample-count) (pcm-bytes->float-pointer buffer buf-len in-bits)))
(let ((frames (quotient sample-count channels)))
(check-ope 'opus-encoder-write
(ope_encoder_write_float (opus-encoder-handle-enc handle) pcm frames))
frames))))
(define (opus-encoder-finish handle)
(dynamic-wind
+17 -7
View File
@@ -1,7 +1,8 @@
(module pcm-converter racket/base
(require ffi/unsafe
"../ffmpeg-definitions.rkt")
"../ffmpeg-definitions.rkt"
"utils.rkt")
(provide pcm-conversion-needed?
make-pcm-converter
@@ -28,7 +29,7 @@
out))
(define (native-signed-ref bs start bytes)
(integer-bytes->integer bs #t (system-big-endian?) start (+ start bytes)))
(int-bytes->integer bs #t (system-big-endian?) start (+ start bytes)))
(define (native-signed-set! bs start bytes value)
(integer->integer-bytes value bytes #t (system-big-endian?) bs start))
@@ -69,21 +70,30 @@
(ptr-set! planes _pointer 0 ptr)
planes))
(define (source-value v source)
(if (eq? v 'source) source v))
(define (target-sample-rate settings input-format)
(source-value
(hash-ref/default settings 'target-sample-rate
(hash-ref/default settings 'sample-rate
(hash-ref input-format 'sample-rate))))
(hash-ref input-format 'sample-rate)))
(hash-ref input-format 'sample-rate)))
(define (target-channels settings input-format)
(source-value
(hash-ref/default settings 'target-channels
(hash-ref/default settings 'channels
(hash-ref input-format 'channels))))
(hash-ref input-format 'channels)))
(hash-ref input-format 'channels)))
(define (target-bits settings input-format)
(let ((source-bits (let ((bits (hash-ref/default input-format 'bits-per-sample 24)))
(if (and (integer? bits) (<= bits 24)) bits 24))))
(source-value
(hash-ref/default settings 'target-bits-per-sample
(hash-ref/default settings 'bits-per-sample
(let ((bits (hash-ref/default input-format 'bits-per-sample 24)))
(if (and (integer? bits) (<= bits 24)) bits 24)))))
(hash-ref/default settings 'bits-per-sample source-bits))
source-bits)))
(define (make-output-format input-format settings)
(let* ((out (copy-hash input-format))