TalkBar voice models β€” ASR download host

First-party download host for the TalkBar desktop app's speech-recognition weights. Every file is a byte-exact repackage of a permissively licensed upstream model (packaging: tar.gz bundles matching the app's install layout, or the upstream file as-is). The app downloads at a pinned commit and verifies SHA-256.

Per-file licenses

File Upstream model Upstream license
parakeet-v3-int8.tar.gz nvidia/parakeet-tdt-0.6b-v3 (int8 ONNX export bundle) CC-BY-4.0
ggml-small.bin OpenAI Whisper small, GGML conversion MIT
whisper-medium-q4_1.bin OpenAI Whisper medium, GGML q4_1 quantization MIT
ggml-large-v3-turbo.bin OpenAI Whisper large-v3-turbo, GGML conversion MIT
ggml-large-v3-q5_0.bin OpenAI Whisper large-v3, GGML q5_0 quantization MIT
giga-am-v3.int8.onnx salute-developers/GigaAM v3, int8 ONNX export MIT
canary-180m-flash.tar.gz nvidia/canary-180m-flash (ONNX bundle) CC-BY-4.0
canary-1b-v2.tar.gz nvidia/canary-1b-v2 (ONNX bundle) CC-BY-4.0
silero_vad.onnx snakers4/silero-vad MIT
voxceleb_resnet34_LM.onnx WeSpeaker voxceleb ResNet34-LM Apache-2.0
diar_streaming_sortformer_4spk-v2.1.onnx nvidia/diar_streaming_sortformer_4spk-v2.1 (ONNX export, no weight changes) NVIDIA Open Model License

Attribution: NVIDIA (Parakeet, Canary), OpenAI (Whisper), SberDevices / salute-developers (GigaAM). No weight modifications beyond quantization / format conversion / archive packaging; all redistribution under the upstream licenses above with attribution.

Integrity

File Bytes SHA-256
parakeet-v3-int8.tar.gz 478,517,071 43d37191602727524a7d8c6da0eef11c4ba24320f5b4730f1a2497befc2efa77
ggml-small.bin 487,601,967 1be3a9b2063867b937e64e2ec7483364a79917e157fa98c5d94b5c1fffea987b
whisper-medium-q4_1.bin 491,852,915 79283fc1f9fe12ca3248543fbd54b73292164d8df5a16e095e2bceeaaabddf57
ggml-large-v3-turbo.bin 1,624,555,275 1fc70f774d38eb169993ac391eea357ef47c88757ef72ee5943879b7e8e2bc69
ggml-large-v3-q5_0.bin 1,081,140,203 d75795ecff3f83b5faa89d1900604ad8c780abd5739fae406de19f23ecd98ad1
giga-am-v3.int8.onnx 224,893,347 2e3fcb7a7b66030336fd10c2fcfb033bd1dc7e1bf238fe5cfd83b1d0cfc9d28e
canary-180m-flash.tar.gz 153,194,652 6d9cfca6118b296e196eaedc1c8fa9788305a7b0f1feafdb6dc91932ab6e53f7
canary-1b-v2.tar.gz 725,374,161 02305b2a25f9cf3e7deaffa7f94df00efa44f442cd55c101c2cb9c000f904666
silero_vad.onnx 1,807,522 a35ebf52fd3ce5f1469b2a36158dba761bc47b973ea3382b3186ca15b1f5af28
voxceleb_resnet34_LM.onnx 26,530,309 7bb2f06e9df17cdf1ef14ee8a15ab08ed28e8d0ef5054ee135741560df2ec068
diar_streaming_sortformer_4spk-v2.1.onnx 492,243,003 f010a62654fd6d2adb70166265df2e53f5a766e354d2cce152b4c7966a04d02d

Document embeddings

bge-m3-q4_k_m.gguf provides BGE-M3 dense embeddings for TalkBar document search. This is an unchanged mirror of the MIT-licensed groonga quantization of BAAI/bge-m3. See provenance and usage and the included MIT license.

File Bytes SHA-256
bge-m3-q4_k_m.gguf 437,778,464 44620fdb0a8d23b5ed57fe5eb8f3842eb9632b4c1363dd602246ac24bd6d1be4

Translation mirrors

Unchanged ONNX runtime packs for the active translation models formerly used from the development Git LFS checkout. Catalog sizes, checksums and inference configuration are preserved. Each pack includes source revisions, original model cards and its declared base-model license.

Downloads last month
61
GGUF
Model size
0.6B params
Architecture
bert
Hardware compatibility
Log In to add your hardware

4-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support