Transcription · Whisper (ggml .bin)
Whisper Large V3 Turbo (default), Large V3, Medium Q8, Medium English Q8 and Small.
Medium English Q8 is for English audio only.
APPLICATION FEATURES
Promic Speech AI is a Windows application for documenting meetings, reviewing what was said and consulting meeting history. In the Professional edition, AI processing and the archive stay on the customer workstation. Archive features depend on the selected usage profile and settings.
Start a free 15-day trial
In Professional, transcription, summaries and queries process content on the Windows workstation. Activation and model downloads use the internet.
02Users, groups and permissions in an organizational archive, AI models larger than 32 GB and AI processing centralized on a server. Deployment by proposal.
03Ask AI questions, request summaries and suggested next steps. Check the cited meetings before acting.
FEATURE COMPARISON
Open a feature for details. Archive steps also depend on the selected usage profile and analysis settings.
Swipe the table to compare editions.
| Feature | Professional | Enterprise | Government |
|---|---|---|---|
| Capture and transcription | |||
Microphones and USB interfacesTranscribe in-person meetings live from an audio device. | Included | Included | Included |
Window audioCapture an online call from a selected window without joining as a bot. | Included | Included | Included |
NDI captureReceive audio and video over the network, such as in chambers and studios. | Included | Included | Included |
Two capture channelsUse two sources at once, each a microphone, a call window or NDI. | Included | Included | Included |
Audio and video filesTranscribe recordings in formats including WAV, MP3, MP4 and WebM. | Included | Included | Included |
MP3 or MP4 recordingRecord MP4 video (1080p, 720p or 360p) or MP3 audio during live transcription; turned on in Settings. | Included | Included | Included |
Local Whisper modelsLarge V3 Turbo is the default; the catalog adds Large V3, Medium and Medium English for English-only audio. | Included | Included | Included |
Language independent of interfaceChoose the spoken language without changing the menu language. | Included | Included | Included |
Noise suppressionAdjust NVIDIA Maxine noise reduction intensity. | Included | Included | Included |
Replacement dictionaryAutomatically correct recurring names, acronyms and terms. | Included | Included | Included |
Reusable text blocksInsert prepared text blocks while editing. | Included | Included | Included |
Automatic savingRecover the document if the computer shuts down mid-session. | Included | Included | Included |
| Speakers | |||
Up to 50 speakers separatedSeparate speech by voice after entering the estimated participant count. | Included | Included | Included |
Voice profilesCreate a profile for each identified voice and save recurring speakers for future meetings. | Included | Included | Included |
Names from what is saidA local model understands phrases such as "I give the floor to John" or "my name is…" and names the passage. | Included | Included | Included |
Participant correctionsRename (updates the whole document), mute (leaves that person out of the text) or drag one speaker onto another to merge them. | Included | Included | Included |
PROMIC chamber integrationIdentify the speaker from the open desk microphone and, optionally, transcribe only during the speaking time. | Not included | By proposal | By proposal |
| Review and documents | |||
Review with playerListen beside the text at 0.25x to 2x speed, with 10 s back and 30 s forward. | Included | Included | Included |
Time linksWith recording on, @hh:mm:ss links in the text take the player to that moment, while the file remains accessible. | Included | Included | Included |
Full text editorFormatting, tables, headers and footers, table of contents, spell check in English, Portuguese and Spanish, password protection and printing. | Included | Included | Included |
Comments and tracked changesReview a document with comments and a record of edits. | Included | Included | Included |
Minutes and session templatesMinutes template with automatic fields, plus chamber events: opening, roll call, voting, result and closing. | Included | Included | Included |
DOCX and PDF exportDOCX from the editor and the archive; PDF from the archive. | Included | Included | Included |
| AI, archive and management | |||
Meeting summariesCreate a summary with a local language model and review it before sharing. | Included | Included | Included |
Qwen and Gemma models for summaries and chatChoose from Qwen and Gemma model families in the catalog, subject to GPU and edition. | Included | Included | Included |
Editable styles and instructionsDetailed, executive minutes, action-focused, chronological by topic, court hearing minutes and more; edit or create styles. | Included | Included | Included |
Document summariesSummarize DOC, DOCX and TXT files outside the archive workflow. | Included | Included | Included |
Decision and reversal extractionExtract the meeting’s decisions and record when a later meeting reverses them. | Included | Included | Included |
Commitment extractionReview suggested actions, owners and due dates; accepted ones become pending items for the next meeting. | Included | Included | Included |
Disagreement extractionRecord unresolved differences and track their outcome. | Included | Included | Included |
Open questions and closeoutOptional: record each meeting’s objective, outcome and the questions left open. | Included | Included | Included |
Connections across meetingsCompare each meeting with earlier ones on the same topic, close old pending items and record reversals. | Included | Included | Included |
Written analysis4 to 8 critical points, why they matter and who should act; checks flag unsourced values and dates and totals that do not add up. | Included | Included | Included |
Decision CenterQueue of what needs attention (overdue, blocked, pending condition, disagreement), a timeline and the statement behind each item. | Included | Included | Included |
Commitment kanbanView by attention or by owner; complete, cancel, reassign or change the due date. | Included | Included | Included |
Meeting archiveBrowse transcripts, summaries and analyses by meeting and topic. | Included | Included | Included |
Chat with sourcesAsk the archive in plain language; citations open the source meeting. | Included | Included | Included |
Archive search and filtersFilter by period and topic; semantic search needs the meetings indexed first. | Included | Included | Included |
| Infrastructure and access | |||
AI in customer infrastructureContent is processed on a workstation or customer server, by edition. | Included | Included | Included |
PostgreSQL archive with pgvectorKeep archive data in customer infrastructure. | Included | Included | Included |
Multiple users and groupsManage users, groups and organizational units. | Not included | Included | Included |
Permissions and visibilityControl actions and meeting access by user, group and unit. | Not included | Included | Included |
Archive on a network serverShare an archive database within the organization. | Not included | Included | Included |
Models larger than 32 GBUse the expanded model catalog with suitable hardware. | Not included | Included | Included |
Centralized server AICentralize GPU and models on a server in the contracted architecture. | Not included | Included | By proposal |
Enterprise and Government are configured by proposal. Government AI processing may run on each workstation or on a server. Integration with a PROMIC chamber system depends on the project.
COMPATIBLE MODELS
Promic Speech AI uses Whisper for transcription and local third-party models for other tasks. In Settings, choose an installed model for transcription, speaker labels, summaries and chat; the in-app catalog offers compatible files for download.
Whisper Large V3 Turbo (default), Large V3, Medium Q8, Medium English Q8 and Small.
Medium English Q8 is for English audio only.
Qwen3.5 4B, 9B, 35B-A3B and 122B-A10B; Qwen3.8 27B; Gemma 4 E4B, 12B, 26B-A4B and 31B; GPT-OSS 20B.
The catalog identifies the role of each variant; for example, Qwen3.5 4B is offered for speaker labels.
Qwen3-Embedding-0.6B Q8_0 and BGE Reranker v2 m3 Q8_0.
Supporting models for indexing, search and result ranking.
These are third-party models downloaded through the in-app catalog. Availability depends on the installed version, NVIDIA GPU, video memory, edition and file variant. Compatibility of external files outside the catalog is not guaranteed.
See the full list and how to choose ↗Transcribe live through a room microphone or USB audio interface without a video meeting platform. For online calls, the application can capture audio from a selected window without adding a meeting bot. You can also open existing audio or video files.
Whisper Large V3 Turbo is the default. The catalog also includes Large V3, Medium Q8 and Medium English Q8, the last of these for English audio. Availability and performance depend on the GPU and installed models.
Set the estimated number of participants before the meeting, and the application separates speech by voice for up to 50 people. This speaker diarization creates a profile for each identified speaker. You can save recurring participants’ voices to recognize them in future meetings, correct assignments and merge duplicate voices. With clear audio, speaker identification is highly accurate; overlapping speech and similar voices may require review.
The transcript appears in an integrated editor. When the recording is available, review can return to the audio passage related to a statement. The reviewed document can be exported as DOCX or PDF.
Choose Qwen, Gemma or GPT-OSS models from the catalog, subject to GPU and edition, and use styles such as executive minutes, detailed, action-focused or chronological by topic. Summary instructions can be edited. Review names, numbers and conclusions before sharing.
For chat, you can also choose GGUF models compatible with your GPU and edition. Ask questions in natural language, request a summary of a meeting or period, and ask for suggested next steps grounded in the archive. Chat consults available transcripts, summaries and facts according to the usage profile, with references to source meetings. Check the sources and review suggestions before acting. Semantic search requires prior indexing.
When fact analysis is enabled, the application organizes suggested decisions, disagreements and commitments. A person reviews and accepts these items; the Decision Center brings together open items and shows context from related meetings.
In Professional, the GPU, AI models and archive database reside on the Windows workstation: meeting content need not be sent to a cloud AI service for transcription, summaries or archive queries. In Enterprise, archive and access can reside in the organization’s infrastructure as defined in the proposal. License activation and initial model downloads use the internet.
An Enterprise deployment connects multiple users to the organization’s archive, with groups and permissions for each role. AI processing can be centralized on a server, with models larger than 32 GB. Capacity and the number of simultaneous meetings are defined in the proposal.