{"id":17468490,"url":"https://github.com/egorsmkv/flashlight-ukrainian","last_synced_at":"2025-06-15T00:35:13.841Z","repository":{"id":60205712,"uuid":"541680849","full_name":"egorsmkv/flashlight-ukrainian","owner":"egorsmkv","description":"The Ukrainian Acoustic Model for Flashlight","archived":false,"fork":false,"pushed_at":"2025-02-20T15:05:14.000Z","size":38,"stargazers_count":7,"open_issues_count":0,"forks_count":0,"subscribers_count":2,"default_branch":"main","last_synced_at":"2025-04-14T21:49:53.811Z","etag":null,"topics":["flashlight","speech-recognition","speech-to-text"],"latest_commit_sha":null,"homepage":"","language":null,"has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"apache-2.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/egorsmkv.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE.txt","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2022-09-26T16:28:10.000Z","updated_at":"2025-02-20T15:05:18.000Z","dependencies_parsed_at":"2025-02-20T15:28:40.186Z","dependency_job_id":"cb230a9f-2bbf-40ef-8913-87c03212042f","html_url":"https://github.com/egorsmkv/flashlight-ukrainian","commit_stats":null,"previous_names":[],"tags_count":1,"template":false,"template_full_name":null,"purl":"pkg:github/egorsmkv/flashlight-ukrainian","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/egorsmkv%2Fflashlight-ukrainian","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/egorsmkv%2Fflashlight-ukrainian/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/egorsmkv%2Fflashlight-ukrainian/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/egorsmkv%2Fflashlight-ukrainian/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/egorsmkv","download_url":"https://codeload.github.com/egorsmkv/flashlight-ukrainian/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/egorsmkv%2Fflashlight-ukrainian/sbom","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":259905430,"owners_count":22929917,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["flashlight","speech-recognition","speech-to-text"],"created_at":"2024-10-18T15:06:41.419Z","updated_at":"2025-06-15T00:35:13.800Z","avatar_url":"https://github.com/egorsmkv.png","language":null,"funding_links":[],"categories":[],"sub_categories":[],"readme":"# Flashlight for Ukrainian\n\n## Community\n\n- Discord: https://bit.ly/discord-uds\n- Speech Recognition: https://t.me/speech_recognition_uk\n- Speech Synthesis: https://t.me/speech_synthesis_uk\n\nSee other Ukrainian models: https://github.com/egorsmkv/speech-recognition-uk\n\n## Overview\n\nThis repository contains the acoustic model for Ukrainian trained on Flashlight framework: https://github.com/flashlight/flashlight/tree/main/flashlight/app/asr\n\n- Architecture: Conformer (300m params)\n- Data in train: Common Voice 10 \u0026 Voice of America\n- Trained epochs: 410\n- Train time: around a week (RTX A4000)\n\n### Download\n\nThe AM model with configuration files published here: https://huggingface.co/Yehor/flashlight-uk/tree/main\n\n## Quality\n\n- WER: 9.0777% (id est the quality is 90.92%)\n- CER: 1.9839%\n\n## How to test?\n\n### Run a container with Flashlight running with CPU\n\n```bash\ndocker-compose up\n\n# and in another termianl\ndocker exec -it flashlight_cpu bash\n```\n\n### Run\n\nJust with an AM:\n\n```\n/root/flashlight/build/bin/asr/fl_asr_test --am /models/uk_am.bin --datadir ''  --emission_dir '' --uselexicon false \\\n --test /data/rows.lst --tokens /models/tokens.txt --lexicon /models/lexicon.txt --show\n ```\n \n With an LM:\n \n ```\n /root/flashlight/build/bin/asr/fl_asr_decode \\\n  --am=/models/uk_am.bin \\\n  --test=/data/labels_absolute.lst \\\n  --maxload=3477 \\\n  --nthread_decoder=2 \\\n  --show \\\n  --showletters \\\n  --lexicon=/models/lexicon.txt \\\n  --uselexicon=false \\\n  --lm=/models/lm_4gram_500k.binary \\\n  --lmtype=kenlm \\\n  --decodertype=wrd \\\n  --beamsize=200 \\\n  --beamsizetoken=200 \\\n  --beamthreshold=20 \\\n  --lmweight=0.75 \\\n  --wordscore=0 \\\n  --eosscore=0 \\\n  --silscore=0 \\\n  --unkscore=0 \\\n  --smearing=max\n ```\n\n- **labels_absolute.lst** is from https://github.com/egorsmkv/cv10-uk-testset-clean\n- **lm_4gram_500k.binary** is from https://huggingface.co/Yehor/kenlm-ukrainian/tree/main/news/lm-4gram-500k\n\n## How to fine-tune on own data?\n\n```\n/root/flashlight/build/bin/asr/fl_asr_train continue /models/ --flagsfile /models/train.flags\n```\n\n`/models/` must contain .bin files\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fegorsmkv%2Fflashlight-ukrainian","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fegorsmkv%2Fflashlight-ukrainian","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fegorsmkv%2Fflashlight-ukrainian/lists"}