{"payload":{"feedbackUrl":"https://github.com/orgs/community/discussions/53140","repo":{"id":551290653,"defaultBranch":"master","name":"ReazonSpeech","ownerLogin":"reazon-research","currentUserCanPush":false,"isFork":false,"isEmpty":false,"createdAt":"2022-10-14T06:16:54.000Z","ownerAvatar":"https://avatars.githubusercontent.com/u/78515509?v=4","public":true,"private":false,"isOrgOwned":true},"refInfo":{"name":"","listCacheKey":"v0:1722501636.0","currentOid":""},"activityList":{"items":[{"before":"fac1f39ae9468793ad6fca319f6ddbf3c6c8482c","after":"6d7fd9a9af6d5d87f60b05f28191c166a17ca5f9","ref":"refs/heads/master","pushedAt":"2024-08-01T08:40:36.000Z","pushType":"push","commitsCount":8,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"ReazonSpeech v2.1.0\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"ReazonSpeech v2.1.0"}},{"before":"4e794a1819b27fa8c1c7b8cfb7ede0045a9475cf","after":null,"ref":"refs/heads/sf/k2","pushedAt":"2024-08-01T06:46:11.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"}},{"before":"2f915c7fbc25dab28b28f8a50b7c639c811b1c0f","after":"4e794a1819b27fa8c1c7b8cfb7ede0045a9475cf","ref":"refs/heads/sf/k2","pushedAt":"2024-08-01T04:19:51.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"Add an experimenal precision mode 'int8-fp32'\n\nThis goes on a middle road of quantization. It uses the int8 models\nfor encoder and joiner, but uses the full precision model for decoder.\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"Add an experimenal precision mode 'int8-fp32'"}},{"before":"4b3b9eda97877f67356b130b001ef88cb7bbf822","after":"2f915c7fbc25dab28b28f8a50b7c639c811b1c0f","ref":"refs/heads/sf/k2","pushedAt":"2024-07-29T03:36:28.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"Show warning if a long audio input is detected\n\nThis is based on the discussion in the upstream. If we feed\na long audio into Zipformer, the performance can degrade\nsignificantly (also causes OOM easily).\n\nAdd an explicit check for input audio lengths.\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"Show warning if a long audio input is detected"}},{"before":"463b08340cf4872881b4fa1d8463d2e95540b58d","after":"4b3b9eda97877f67356b130b001ef88cb7bbf822","ref":"refs/heads/sf/k2","pushedAt":"2024-07-28T13:04:47.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"Show warning if a long audio input is detected\n\nIf we feed a long audio to Zipformer, it easily exhauts the\nGPU memory. Let's add a cautionary warning.\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"Show warning if a long audio input is detected"}},{"before":null,"after":"463b08340cf4872881b4fa1d8463d2e95540b58d","ref":"refs/heads/sf/k2","pushedAt":"2024-07-28T09:10:42.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"Describe the new K2 module in README.rst\n\nThis adds the link to 'pkg/k2-asr', so that users can navigate\nto that subpackage.\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"Describe the new K2 module in README.rst"}},{"before":"e5b2e303329e346c82c07d116b1f2b8eb61c5168","after":"fac1f39ae9468793ad6fca319f6ddbf3c6c8482c","ref":"refs/heads/master","pushedAt":"2024-07-17T04:47:39.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"Add a new Python package 'reazonspeech.k2.asr'\n\nThis adds the support for a fast lightweight ASR model based on K2.\n\n https://github.com/k2-fsa/k2\n\nThe pre-trained model is available in the ONNX format (in particular,\nwe make use of \"sherpa_onnx\"), so it can be used on platforms such\nas ARM64.\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"Add a new Python package 'reazonspeech.k2.asr'"}},{"before":"ef9033792609bf62ed1c7eb5041201486e6fd7a0","after":"e5b2e303329e346c82c07d116b1f2b8eb61c5168","ref":"refs/heads/master","pushedAt":"2024-04-01T02:29:56.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"Avoid returning whitespace token from SentencePiece\n\nBased on patch from Yuta Sasaki.\n\nSentencePiece treats whitespace as a normal token. For this reason,\nNeMo sometimes includes meta tokens (U+2581) in the inference results.\n\nThis evidently ended up confusing decode_hypothesis() to produce fuzzy\ntimestamps for some segments. This patch fixes it.\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"Avoid returning whitespace token from SentencePiece"}},{"before":"9170995bcd0af188c75f9d58100fb627ac1eb2af","after":"ef9033792609bf62ed1c7eb5041201486e6fd7a0","ref":"refs/heads/master","pushedAt":"2024-04-01T02:28:25.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"ReazonSpeech v2.0.1\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"ReazonSpeech v2.0.1"}},{"before":"481efa9cbb91dc8f1252796d5dd7ef29d989933c","after":"9170995bcd0af188c75f9d58100fb627ac1eb2af","ref":"refs/heads/master","pushedAt":"2024-02-27T07:10:12.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"Allow users to access Hypothesis objects\n\nThis adds a new configuration option 'raw_hypothesis' that teaches\nReazonSpeech to return NeMo's Hypothesis object (default is False).\n\nThis object contains gory details about the decoding state, and\nseems to be useful for advanced users.\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"Allow users to access Hypothesis objects"}},{"before":"ee81069a6b6c253839778b8d1e69835450b33e0a","after":"481efa9cbb91dc8f1252796d5dd7ef29d989933c","ref":"refs/heads/master","pushedAt":"2024-02-22T06:24:52.000Z","pushType":"push","commitsCount":2,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"Merge 'dev/segments-time' of yuta0306/ReazonSpeech\n\nPatch from Sasaki Yuta.\n\nA bit of testing suggests that the \"step\" timings refer to the beginning\nof audio frames. For this reason, we needed add 80ms to the end of the\nsegment in order not to miss the last phoneme. Fix it thusly.","shortMessageHtmlLink":"Merge 'dev/segments-time' of yuta0306/ReazonSpeech"}},{"before":"3cb0462672ec7b2ac50d7670ee0b2999be5ebdc0","after":null,"ref":"refs/heads/sf/win32","pushedAt":"2024-02-17T05:03:12.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"}},{"before":"e4590baeba669f577723498b6b57b4abf3f7d4b2","after":"ee81069a6b6c253839778b8d1e69835450b33e0a","ref":"refs/heads/master","pushedAt":"2024-02-17T05:02:38.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"nemo-asr: add create_tempfile() for Win32 compat\n\nWindows locks open files. For this reason, a file path created by\nNamedTemporaryFile cannot be reopened on Windows.\n\nThis patch tries to mitigate the issue by:\n\n 1. Switch to manual clean-up using custom context manager.\n 2. Close a file handler before passing it to NeMo.\n\nConfirmed to work on Windows Server 2022 with Python 3.11.8.\n\nNote: Read https://bugs.python.org/issue14243 for related discussion.\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"nemo-asr: add create_tempfile() for Win32 compat"}},{"before":"327db557584cddbdafa2f165075744592dee9a49","after":"3cb0462672ec7b2ac50d7670ee0b2999be5ebdc0","ref":"refs/heads/sf/win32","pushedAt":"2024-02-17T04:39:53.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"nemo-asr: add create_tempfile() for Win32 compat\n\nWindows locks open files. For this reason, a file path created by\nNamedTemporaryFile cannot be reopened on Windows.\n\nThis patch tries to mitigate the issue by:\n\n 1. Switch to manual clean-up using custom context manager.\n 2. Close a file handler before passing it to NeMo.\n\nRead https://bugs.python.org/issue14243 for related discussion.\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"nemo-asr: add create_tempfile() for Win32 compat"}},{"before":"186302e34a4791507945de6e93f13fbee497eb98","after":"327db557584cddbdafa2f165075744592dee9a49","ref":"refs/heads/sf/win32","pushedAt":"2024-02-16T16:35:05.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"Fix file clean-up for Windows\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"Fix file clean-up for Windows"}},{"before":null,"after":"186302e34a4791507945de6e93f13fbee497eb98","ref":"refs/heads/sf/win32","pushedAt":"2024-02-16T15:46:15.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"nemo-asr: add create_tempfile() for Win32 compat\n\nWindows locks open files. For this reason, a file path created by\nNamedTemporaryFile cannot be reopened on Windows.\n\nThis patch tries to mitigate the issue by:\n\n 1. Switch to manual clean-up by setting delete flag on Windows.\n 2. Close a file handler before passing it to NeMo.\n\nRead https://bugs.python.org/issue14243 for upstream discussion.\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"nemo-asr: add create_tempfile() for Win32 compat"}},{"before":"42eeef6c3b76d745ccb94c6e07b8ac24659f6032","after":null,"ref":"refs/heads/sf/colab-en","pushedAt":"2024-02-16T13:26:39.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"}},{"before":"a98123d50b646bcad113508c4dfc068157ab4a97","after":"e4590baeba669f577723498b6b57b4abf3f7d4b2","ref":"refs/heads/master","pushedAt":"2024-02-16T10:20:04.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"Translate Google Colab note into English\n\nThis is a translation of 'ReazonSpeech_v2_0_en.ipynb', which I\noriginally write in Japanese.\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"Translate Google Colab note into English"}},{"before":"83cdbd21535ee6e5d1b2d95eba8d9f967cb3d934","after":"42eeef6c3b76d745ccb94c6e07b8ac24659f6032","ref":"refs/heads/sf/colab-en","pushedAt":"2024-02-16T10:15:43.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"More fixes","shortMessageHtmlLink":"More fixes"}},{"before":"225b9adbe21d1d5a8b8fa8cb5fa716be9e3ab5b8","after":"83cdbd21535ee6e5d1b2d95eba8d9f967cb3d934","ref":"refs/heads/sf/colab-en","pushedAt":"2024-02-16T10:09:16.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"Polish up wording in 'ReazonSpeech_v2_0_en.ipynb'","shortMessageHtmlLink":"Polish up wording in 'ReazonSpeech_v2_0_en.ipynb'"}},{"before":"8c24b42ad356580c7f2bc545459d72565271b4b2","after":"225b9adbe21d1d5a8b8fa8cb5fa716be9e3ab5b8","ref":"refs/heads/sf/colab-en","pushedAt":"2024-02-16T10:01:58.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"Fix up English translation","shortMessageHtmlLink":"Fix up English translation"}},{"before":null,"after":"8c24b42ad356580c7f2bc545459d72565271b4b2","ref":"refs/heads/sf/colab-en","pushedAt":"2024-02-16T09:49:23.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"Translate Google Colab note into English\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"Translate Google Colab note into English"}},{"before":"da898079d87bf36671571a0ca2700b0279608196","after":"a98123d50b646bcad113508c4dfc068157ab4a97","ref":"refs/heads/master","pushedAt":"2024-02-16T09:26:16.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"espnet-asr: add Python 3.8 compatibility\n\nThe type hinting syntax is available since Python 3.9. For this\nreason, it ended up causing an execption on Python 3.8.\n\n TypeError: 'type' object is not subscriptable\n\nFix it thusly. Tested on Ubuntu 20.04.\n\nNote: This patch is only for ESPnet module. Since the oldest Python\nversion nvidia/nemo supports is 3.10, pkg/nemo-asr does not support\nthe older Python versions.\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"espnet-asr: add Python 3.8 compatibility"}},{"before":"d5f7455010becdf5573c79426f9e67e81f441e54","after":"da898079d87bf36671571a0ca2700b0279608196","ref":"refs/heads/master","pushedAt":"2024-02-15T09:11:34.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"Add instruction about Google Colab GPU\n\nGoogle Colab chooses a CPU instance by default. Let's add some\ninstruction about GPU usage.\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"Add instruction about Google Colab GPU"}},{"before":"b9b5dd6e604c2c59d31314f98aac9638891fee83","after":null,"ref":"refs/heads/sf/colab","pushedAt":"2024-02-15T07:30:41.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"}},{"before":"c3b7d2a471dd393b95a10d3945f35e4e0d9753f0","after":"d5f7455010becdf5573c79426f9e67e81f441e54","ref":"refs/heads/master","pushedAt":"2024-02-15T07:29:58.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"Add a tutorial Google Colab file\n\nThis adds an iPython notebook file that can be used with Google\nColab.\n\nThis should allow users to experiment with our v2.0 model.\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"Add a tutorial Google Colab file"}},{"before":null,"after":"b9b5dd6e604c2c59d31314f98aac9638891fee83","ref":"refs/heads/sf/colab","pushedAt":"2024-02-15T06:39:13.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"Add a tutorial Google Colab file\n\nThis adds an iPython notebook file that can be used with Google\nColab.\n\nThis should allow users to experiment with our v2.0 model.\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"Add a tutorial Google Colab file"}},{"before":"1bd59a9a3cda240c1040fcfd8aec2e42f0d34146","after":"c3b7d2a471dd393b95a10d3945f35e4e0d9753f0","ref":"refs/heads/master","pushedAt":"2024-02-13T16:41:25.000Z","pushType":"push","commitsCount":10,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"nemo-asr: Reduce PAD_SECDONDS from 1.0 to 0.5\n\nThis makes the inference speed 15% faster, while keeping the decoding\naccuracy. Here is the comparison:\n\n # BEFORE AFTER\n just 7.23 7.31\n cv 9.00 8.81\n tedx 10.11 10.42\n book 17.26 17.22\n RTF 0.60 0.51\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"nemo-asr: Reduce PAD_SECDONDS from 1.0 to 0.5"}},{"before":"30587ade7b3a770fc988752eb866b7464dec7b50","after":null,"ref":"refs/heads/next","pushedAt":"2024-02-05T07:32:38.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"}},{"before":"f4e356188b86509176abf8932c65b7b14595c76e","after":"1bd59a9a3cda240c1040fcfd8aec2e42f0d34146","ref":"refs/heads/master","pushedAt":"2024-02-05T05:09:33.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"fujimotos","name":"Fujimoto Seiji","path":"/fujimotos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8974561?s=80&v=4"},"commit":{"message":"Revert \"Override sentencepiece version to >= v0.1.98\"\n\nThis reverts commit f4e356188b86509176abf8932c65b7b14595c76e.\n\nIt turns out that this confuses setuptools's requirement resolution,\ncausing `reazonspeech` command to fail with dependency exception\n(confirmed with ESPnet v.202310).\n\nWe need to wait for the upstream project to merge espnet#5532.\n\nSigned-off-by: Fujimoto Seiji ","shortMessageHtmlLink":"Revert \"Override sentencepiece version to >= v0.1.98\""}}],"hasNextPage":true,"hasPreviousPage":false,"activityType":"all","actor":null,"timePeriod":"all","sort":"DESC","perPage":30,"cursor":"djE6ks8AAAAEjuZiwQA","startCursor":null,"endCursor":null}},"title":"Activity ยท reazon-research/ReazonSpeech"}