whispercpp 1.3.6 → 1.3.8

data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm.wgsl

@@ -1,123 +0,0 @@
-#define(VARIANTS)
-
-[
-  {
-    "DECLS": ["NOT_INPLACE"]
-  },
-  {
-    "SHADER_SUFFIX": "inplace",
-    "DECLS": ["INPLACE"]
-  },
-]
-
-#end(VARIANTS)
-
-#define(DECLS)
-
-#decl(NOT_INPLACE)
-
-fn update(src_offset: u32, dst_offset: u32, scale: f32) {
-    dst[dst_offset] = scale * src[src_offset];
-}
-
-@group(0) @binding(1)
-var<storage, read_write> dst: array<f32>;
-
-@group(0) @binding(2)
-var<uniform> params: Params;
-
-#enddecl(NOT_INPLACE)
-
-#decl(INPLACE)
-
-fn update(src_offset: u32, dst_offset: u32, scale: f32) {
-    src[dst_offset] = scale * src[src_offset];
-}
-
-@group(0) @binding(1)
-var<uniform> params: Params;
-
-#enddecl(INPLACE)
-
-#end(DECLS)
-
-#define(SHADER)
-
-struct Params {
-    offset_src: u32, // in elements
-    offset_dst: u32, // in elements
-
-    // Strides (in elements)
-    stride_src1: u32,
-    stride_src2: u32,
-    stride_src3: u32,
-
-    stride_dst1: u32,
-    stride_dst2: u32,
-    stride_dst3: u32,
-
-    // Shape of src/dst
-    ne0: u32,
-    ne1: u32,
-    ne2: u32,
-    ne3: u32,
-
-    eps: f32
-};
-
-@group(0) @binding(0)
-var<storage, read_write> src: array<f32>;
-
-DECLS
-
-override wg_size: u32;
-var<workgroup> scratch: array<f32, wg_size>;
-
-@compute @workgroup_size(wg_size)
-fn main(@builtin(workgroup_id) wid: vec3<u32>,
-        @builtin(local_invocation_id) lid: vec3<u32>) {
-
-    // one thread per row
-    var i = wid.x;
-    let i3 = i / (params.ne2 * params.ne1);
-    i = i % (params.ne2 * params.ne1);
-    let i2 = i / params.ne1;
-    let i1 = i % params.ne1;
-    let i_src_row = params.offset_src + i3 * params.stride_src3 + i2 * params.stride_src2 + i1 * params.stride_src1;
-    let i_dst_row = params.offset_dst + i3 * params.stride_dst3 + i2 * params.stride_dst2 + i1 * params.stride_dst1;
-
-    let elems = (params.ne0 + wg_size - 1) / wg_size;
-
-    var sum = 0.0f;
-    var col = lid.x;
-    for (var j: u32 = 0; j < elems; j++) {
-        if (col >= params.ne0) {
-            break;
-        }
-        sum += pow(src[i_src_row + col], 2.0);
-        col += wg_size;
-    }
-
-    scratch[lid.x] = sum;
-    workgroupBarrier();
-    var offset = wg_size / 2;
-    while (offset > 0) {
-        if (lid.x < offset) {
-            scratch[lid.x] += scratch[lid.x + offset];
-        }
-        offset = offset / 2;
-        workgroupBarrier();
-    }
-    sum = scratch[0];
-
-    let scale = 1.0/sqrt(sum/f32(params.ne0) + params.eps);
-    col = lid.x;
-    for (var j: u32 = 0; j < elems; j++) {
-        if (col >= params.ne0) {
-            break;
-        }
-        update(i_src_row + col, i_dst_row + col, scale);
-        col += wg_size;
-    }
-}
-#end(SHADER)

data/ext/sources/tests/CMakeLists.txt

@@ -1,112 +0,0 @@
-set(CMAKE_CXX_STANDARD 17)
-set(CMAKE_CXX_STANDARD_REQUIRED ON)
-
-if (EMSCRIPTEN)
-    #
-    # test-whisper-js
-
-    set(TEST_TARGET test-whisper-js)
-
-    add_test(NAME ${TEST_TARGET}
-        COMMAND node test-whisper.js --experimental-wasm-threads
-        WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
-        )
-
-    return()
-endif()
-
-set(TEST_TARGET test-whisper-cli-tiny)
-add_test(NAME ${TEST_TARGET}
-    COMMAND $<TARGET_FILE:whisper-cli>
-    -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-tiny.bin -l fr
-    -f ${PROJECT_SOURCE_DIR}/samples/jfk.wav)
-set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "tiny;gh")
-
-set(TEST_TARGET test-whisper-cli-tiny.en)
-add_test(NAME ${TEST_TARGET}
-    COMMAND $<TARGET_FILE:whisper-cli>
-    -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-tiny.en.bin
-    -f ${PROJECT_SOURCE_DIR}/samples/jfk.wav)
-set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "tiny;en")
-
-set(TEST_TARGET test-whisper-cli-base)
-add_test(NAME ${TEST_TARGET}
-    COMMAND $<TARGET_FILE:whisper-cli>
-    -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-base.bin -l fr
-    -f ${PROJECT_SOURCE_DIR}/samples/jfk.wav)
-set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "base")
-
-set(TEST_TARGET test-whisper-cli-base.en)
-add_test(NAME ${TEST_TARGET}
-    COMMAND $<TARGET_FILE:whisper-cli>
-    -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-base.en.bin
-    -f ${PROJECT_SOURCE_DIR}/samples/jfk.wav)
-set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "base;en")
-
-set(TEST_TARGET test-whisper-cli-small)
-add_test(NAME ${TEST_TARGET}
-    COMMAND $<TARGET_FILE:whisper-cli>
-    -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-small.bin -l fr
-    -f ${PROJECT_SOURCE_DIR}/samples/jfk.wav)
-set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "small")
-
-set(TEST_TARGET test-whisper-cli-small.en)
-add_test(NAME ${TEST_TARGET}
-    COMMAND $<TARGET_FILE:whisper-cli>
-    -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-small.en.bin
-    -f ${PROJECT_SOURCE_DIR}/samples/jfk.wav)
-set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "small;en")
-
-set(TEST_TARGET test-whisper-cli-medium)
-add_test(NAME ${TEST_TARGET}
-    COMMAND $<TARGET_FILE:whisper-cli>
-    -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-medium.bin -l fr
-    -f ${PROJECT_SOURCE_DIR}/samples/jfk.wav)
-set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "medium")
-
-set(TEST_TARGET test-whisper-cli-medium.en)
-add_test(NAME ${TEST_TARGET}
-    COMMAND $<TARGET_FILE:whisper-cli>
-    -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-medium.en.bin
-    -f ${PROJECT_SOURCE_DIR}/samples/jfk.wav)
-set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "medium;en")
-
-set(TEST_TARGET test-whisper-cli-large)
-add_test(NAME ${TEST_TARGET}
-    COMMAND $<TARGET_FILE:whisper-cli>
-    -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-large.bin
-    -f ${PROJECT_SOURCE_DIR}/samples/jfk.wav)
-set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "large")
-
-if (WHISPER_FFMPEG)
-    set(TEST_TARGET test-whisper-cli-tiny-mp3)
-    # Check with reviewers: any way to check the output transcription via ctest (diff, ...)?
-    add_test(NAME ${TEST_TARGET}
-      COMMAND $<TARGET_FILE:whisper-cli>
-      -m ${PROJECT_SOURCE_DIR}/models/for-tests-ggml-tiny.en.bin
-      -f ${PROJECT_SOURCE_DIR}/samples/jfk.mp3)
-    set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "tiny;mp3")
-endif()
-
-# VAD test tests VAD in isolation
-set(VAD_TEST test-vad)
-add_executable(${VAD_TEST} ${VAD_TEST}.cpp)
-target_include_directories(${VAD_TEST} PRIVATE ../include ../ggml/include ../examples)
-target_link_libraries(${VAD_TEST} PRIVATE common)
-target_compile_definitions(${VAD_TEST} PRIVATE
-    VAD_MODEL_PATH="${PROJECT_SOURCE_DIR}/models/for-tests-silero-v6.2.0-ggml.bin"
-    SAMPLE_PATH="${PROJECT_SOURCE_DIR}/samples/jfk.wav")
-add_test(NAME ${VAD_TEST} COMMAND ${VAD_TEST})
-set_tests_properties(${VAD_TEST} PROPERTIES LABELS "unit")
-
-# VAD test full uses whisper_full with VAD enabled
-set(VAD_TEST test-vad-full)
-add_executable(${VAD_TEST} ${VAD_TEST}.cpp)
-target_include_directories(${VAD_TEST} PRIVATE ../include ../ggml/include ../examples)
-target_link_libraries(${VAD_TEST} PRIVATE common)
-target_compile_definitions(${VAD_TEST} PRIVATE
-    WHISPER_MODEL_PATH="${PROJECT_SOURCE_DIR}/models/ggml-base.en.bin"
-    VAD_MODEL_PATH="${PROJECT_SOURCE_DIR}/models/for-tests-silero-v6.2.0-ggml.bin"
-    SAMPLE_PATH="${PROJECT_SOURCE_DIR}/samples/jfk.wav")
-add_test(NAME ${VAD_TEST} COMMAND ${VAD_TEST})
-set_tests_properties(${VAD_TEST} PROPERTIES LABELS "base;en")

data/ext/sources/tests/earnings21/eval.mk

@@ -1,58 +0,0 @@
-PYTHON = python
-
-WHISPER_PREFIX = ../../
-WHISPER_MODEL = tiny
-
-WHISPER_CLI = $(WHISPER_PREFIX)build/bin/whisper-cli
-WHISPER_FLAGS = --no-prints --language en --output-txt
-
-# You can create eval.conf to override the WHISPER_* variables
-# defined above.
--include eval.conf
-
-# Add  `EARNINGS21_EVAL10 = yes` to eval.conf to switch to a
-# 10-hour subset. See "speech-datasets/earnings21/README.md" for
-# more details about this subset.
-ifdef EARNINGS21_EVAL10
-METADATA_CSV = speech-datasets/earnings21/eval10-file-metadata.csv
-AUDIO_SRCS = speech-datasets/earnings21/media/4320211.mp3 \
-             speech-datasets/earnings21/media/4341191.mp3 \
-             speech-datasets/earnings21/media/4346818.mp3 \
-             speech-datasets/earnings21/media/4359971.mp3 \
-             speech-datasets/earnings21/media/4365024.mp3 \
-             speech-datasets/earnings21/media/4366522.mp3 \
-             speech-datasets/earnings21/media/4366893.mp3 \
-             speech-datasets/earnings21/media/4367535.mp3 \
-             speech-datasets/earnings21/media/4383161.mp3 \
-             speech-datasets/earnings21/media/4384964.mp3 \
-             speech-datasets/earnings21/media/4387332.mp3
-else
-METADATA_CSV = speech-datasets/earnings21/earnings21-file-metadata.csv
-AUDIO_SRCS = $(sort $(wildcard speech-datasets/earnings21/media/*.mp3))
-endif
-
-TRANS_TXTS = $(addsuffix .txt, $(AUDIO_SRCS))
-
-# We output the evaluation result to this file.
-DONE = $(WHISPER_MODEL).txt
-
-all: $(DONE)
-
-$(DONE): $(TRANS_TXTS)
-	$(PYTHON) eval.py $(METADATA_CSV) > $@.tmp
-	mv $@.tmp $@
-
-# Note: This task writes to a temporary file first to
-# create the target file atomically.
-%.mp3.txt: %.mp3
-	$(WHISPER_CLI) $(WHISPER_FLAGS) --model $(WHISPER_PREFIX)models/ggml-$(WHISPER_MODEL).bin --file $^ --output-file $^.tmp
-	mv $^.tmp.txt $^.txt
-
-archive:
-	tar -czf $(WHISPER_MODEL).tar.gz --exclude="*.mp3" speech-datasets/earnings21/media $(DONE)
-
-clean:
-	@rm -f $(TRANS_TXTS)
-	@rm -f $(DONE)
-
-.PHONY: all archive clean

data/ext/sources/tests/earnings21/eval.py

@@ -1,68 +0,0 @@
-import os
-import sys
-import glob
-import jiwer
-from normalizers import EnglishTextNormalizer
-
-def decode_hypothesis(b):
-    try:
-        # Depending on platforms, Whisper can emit a left double quotation
-        # mark (0x93), which is Microsoft's extension to ASCII. See #3185
-        # for the background.
-        return b.decode('windows-1252')
-    except UnicodeDecodeError:
-        return b.decode('utf-8', errors='ignore')
-
-def get_reference():
-    ref = {}
-    for path in glob.glob("speech-datasets/earnings21/transcripts/nlp_references/*.nlp"):
-        code = os.path.basename(path).replace(".nlp", "")
-        buf = []
-        with open(path) as fp:
-            fp.readline()
-            for line in fp:
-                token = line.split("|", maxsplit=1)[0]
-                buf.append(token)
-            ref[code] = " ".join(buf)
-    return ref
-
-def get_hypothesis():
-    hyp = {}
-    for path in glob.glob("speech-datasets/earnings21/media/*.mp3.txt"):
-        with open(path, 'rb') as fp:
-            text = decode_hypothesis(fp.read()).strip()
-        code = os.path.basename(path).replace(".mp3.txt", "")
-        hyp[code] = text
-    return hyp
-
-def get_codes(metadata_csv):
-    codes = []
-    with open(metadata_csv) as fp:
-        fp.readline()
-        for line in fp:
-            codes.append(line.split(",")[0])
-    return sorted(codes)
-
-def main():
-    if len(sys.argv) < 2:
-        print("Usage: %s METADATA_CSV" % sys.argv[0], file=sys.stderr)
-        return 1
-
-    metadata_csv = sys.argv[1]
-    normalizer = EnglishTextNormalizer()
-
-    ref_orig = get_reference()
-    hyp_orig = get_hypothesis()
-
-    ref_clean = []
-    hyp_clean = []
-
-    for code in get_codes(metadata_csv):
-        ref_clean.append(normalizer(ref_orig[code]))
-        hyp_clean.append(normalizer(hyp_orig[code]))
-
-    wer = jiwer.wer(ref_clean, hyp_clean)
-    print(f"WER: {wer * 100:.2f}%")
-
-if __name__ == "__main__":
-    main()

data/ext/sources/tests/earnings21/normalizers/__init__.py

@@ -1,2 +0,0 @@
-from .basic import BasicTextNormalizer as BasicTextNormalizer
-from .english import EnglishTextNormalizer as EnglishTextNormalizer

data/ext/sources/tests/earnings21/normalizers/basic.py

@@ -1,80 +0,0 @@
-import re
-import unicodedata
-
-import regex
-
-# non-ASCII letters that are not separated by "NFKD" normalization
-ADDITIONAL_DIACRITICS = {
-    "œ": "oe",
-    "Œ": "OE",
-    "ø": "o",
-    "Ø": "O",
-    "æ": "ae",
-    "Æ": "AE",
-    "ß": "ss",
-    "ẞ": "SS",
-    "đ": "d",
-    "Đ": "D",
-    "ð": "d",
-    "Ð": "D",
-    "þ": "th",
-    "Þ": "th",
-    "ł": "l",
-    "Ł": "L",
-}
-
-
-def remove_symbols_and_diacritics(s: str, keep=""):
-    """
-    Replace any other markers, symbols, and punctuations with a space,
-    and drop any diacritics (category 'Mn' and some manual mappings)
-    """
-    return "".join(
-        (
-            c
-            if c in keep
-            else (
-                ADDITIONAL_DIACRITICS[c]
-                if c in ADDITIONAL_DIACRITICS
-                else (
-                    ""
-                    if unicodedata.category(c) == "Mn"
-                    else " " if unicodedata.category(c)[0] in "MSP" else c
-                )
-            )
-        )
-        for c in unicodedata.normalize("NFKD", s)
-    )
-
-
-def remove_symbols(s: str):
-    """
-    Replace any other markers, symbols, punctuations with a space, keeping diacritics
-    """
-    return "".join(
-        " " if unicodedata.category(c)[0] in "MSP" else c
-        for c in unicodedata.normalize("NFKC", s)
-    )
-
-
-class BasicTextNormalizer:
-    def __init__(self, remove_diacritics: bool = False, split_letters: bool = False):
-        self.clean = (
-            remove_symbols_and_diacritics if remove_diacritics else remove_symbols
-        )
-        self.split_letters = split_letters
-
-    def __call__(self, s: str):
-        s = s.lower()
-        s = re.sub(r"[<\[][^>\]]*[>\]]", "", s)  # remove words between brackets
-        s = re.sub(r"\(([^)]+?)\)", "", s)  # remove words between parenthesis
-        s = self.clean(s).lower()
-
-        if self.split_letters:
-            s = " ".join(regex.findall(r"\X", s, regex.U))
-
-        s = re.sub(
-            r"\s+", " ", s
-        )  # replace any successive whitespace characters with a space
-
-        return s