Speech Insights: assisted review with timestamped evidence
Turn authorized transcripts into cited findings and candidate agreements or tasks, subject to human review.
Need a transcript? Open the existing module with a recording you are authorized to process. When it finishes, return to Speech Insights through Solutions to select the source. Analysis reuses its text without resending audio to the GPU.
Open Speech to Text uploadFrom an authorized source to a reviewable finding
1. Select a completed transcript
Import a source from your own account into a private project. The server checks authorization, completion and limits, then keeps a snapshot with hashes and timestamped segments. Times are not fabricated for incompatible legacy formats.
2. Define a literal rubric
Configure versioned criteria and terms. A rule matches any or all of its terms within the same segment. It works without an AI provider and does not switch automatically to Kungen.
3. Review each finding
Inspect citations and their intervals, then accept, reject or request changes. Download JSON or TXT evidence with its review. Suggestions still require human judgment before use.
Current service configuration
An enabled capability describes published configuration; it does not grant access, demonstrate provider health or guarantee an operation succeeds. The dashboard checks account, project, source, validity, budget and engine before admission.
Import your completed transcripts
Checking…Literal rule matches within each segment
Checking…Configurable, versioned rubrics
Checking…Analysis with a recoverable reference
Checking…Versioned human review
Checking…Private JSON and TXT exports
Checking…ViciousByte evidence signature
Checking…Optional analysis through Kungen
Checking…
How to interpret the result
Literal rules: points count observed criteria, not the semantic quality of a conversation. An unobserved term does not establish universal absence. Citations are representative, not exhaustive coverage of every occurrence.
Agreements and tasks are candidates derived from explicit rubric criteria, or suggestions from the optional engine. They do not execute actions, assign owners or certify commitments automatically.
Timestamped evidence: each quote is checked against its source segment and times. This does not establish speaker identity, the truth of a statement or the correctness of model interpretation. Synchronized audio playback is not offered.
When configured, a ViciousByte evidence signature attests to application report integrity; it does not certify transcription accuracy, compliance or human acceptance.
Kungen is optional and requires consent
Before choosing Kungen, the dashboard presents the current notice identifying the configured gateway and model. You must explicitly accept that notice to send text, segments and rubric; audio is not sent. The gateway may involve external providers: data is not promised to stay within a private perimeter or avoid external logs.
If confirmation of a provider request is lost, analysis keeps its reference and requires reconciliation. It is not resubmitted automatically. A cancellation request is not confirmed cancellation and does not prove the provider did not process the content.
Source, review and analysis limits
Analysis has a separate budget from transcription minutes: ASR minutes are not charged again and no new GPU job is created. A started attempt may consume a unit even if it fails, returns an invalid result or has an uncertain provider outcome. Daily admission counters are not refunded by cancellation; the dashboard shows the applicable quota.
Published limits could not be confirmed. Check the dashboard; available quota, capacity or retention is not assumed.
Private sources and retention
Projects belong to one account. Private snapshots of text and segments are stored with evidence, rubrics and reviews. Reading and exporting require authorization. Original audio and the existing transcript follow their own retention flow; Speech Insights does not copy audio or offer guaranteed automatic anonymization.
Deleting a source, project or account, or reaching retention expiry, revokes the corresponding access and queues snapshot, evidence and related payload purging. Physical deletion is asynchronous. Minimal quota and idempotency receipts are retained separately; an uncertain external result requires reconciliation.
A current transcription subscription may enable analysis when confirmed by the server, with a separate budget. A pilot grant or paid Speech Insights period may also enable it for the project. Purchasing Speech Insights does not add ASR minutes or create GPU jobs. A pending request grants no permissions and starts no payment.
Plans and additional units
Prepaid 30-day base per project and service. Manual renewal after expiry; no automatic charges or automatic overage purchases.
Checking commercial terms…
Proposed pricing
A public view of a commercial proposal that has not been approved. These amounts are not available purchase offers and do not enable charges. Final terms may change after review.
Proposal dated October 25, 2026 · USD · base amounts exclude VAT (16%) · 30 days · manual renewal · no automatic chargesStarter
USD 15.00
20 analysis attempts
Project base for one 30-day periodPro
USD 39.00
100 analysis attempts
Project base for one 30-day periodBusiness
USD 99.00
300 analysis attempts
Project base for one 30-day period20 analysis attempts
USD 15.00 · Proposed · subject to review
60 analysis attempts
USD 39.00 · Proposed · subject to review
120 analysis attempts
USD 72.00 · Proposed · subject to review
Request pilot access
Register a Speech Insights request for your account. This does not create a subscription, activate permissions or execute a payment.