Documentation
Audionyx is an audio conversation intelligence platform. It turns spoken media — podcasts, interviews, debates, meetings — into structured, queryable intelligence.
Audionyx is currently in public beta. Some features may be incomplete or subject to change. API access documentation will be published soon.
Getting Started
From sign-up to your first analysis in under five minutes.
- 1
Create an account
Sign up for free. Free accounts let you browse and view public reports. Upgrade to Pro to start ingesting and analyzing audio.
- 2
Subscribe to a paid plan
Go to Billing and choose Pro ($14/mo) or Newsroom ($39/seat/mo). Pro gives you 20 jobs/month; Newsroom gives teams a shared pool of 100 jobs/month.
- 3
Submit audio
Go to Dashboard → New Ingestion. Paste a public URL (podcast episode, direct audio link, etc.) or upload a file. Select the content type and optionally add trusted reference URLs for fact-checking.
- 4
Wait for processing
The 7-stage pipeline runs automatically. A typical 60-minute episode processes in 3–8 minutes. You'll see live progress on the job page.
- 5
Explore results
Once complete, the Analysis tab shows topics, claims with verdicts, sentiment, and a full transcript. Use the Q&A tab to ask questions in plain language.
- 6
Share or keep private
Publish your analysis as a public report with a custom URL, or keep it private to your account.
Features
What Audionyx extracts from every audio file.
Transcription & Speaker Diarization
Full verbatim transcripts with per-speaker attribution. Supports up to 10 distinct speakers. Speaker names are editable after processing.
Topic Extraction
Automatically identifies the key topics discussed, grouped by segment. Each topic includes a summary and relevance score.
Claim Detection & Fact-Checking
Extracts factual claims from the transcript and verifies each one against live web sources. Returns a verdict with confidence score and cited evidence.
Sentiment Analysis
Tracks sentiment across the conversation over time — per speaker and per segment — revealing tone shifts and emotional arcs.
Conversational Q&A
Ask plain-language questions about any audio file. Responses are grounded in the transcript and cited to specific timestamps.
Public Sharing
Publish any analysis as a public report with a custom slug. Share a link to a fully interactive, SEO-indexed page.
Explore Library
Browse publicly shared analyses from other users. Filter by type (podcast, debate, interview), sort by credibility or claim count.
How Fact-Checking Works
Claim verification runs automatically as part of the pipeline. Here is exactly what happens for each claim.
Claim extraction
Claude identifies factual assertions in the transcript — statements that are verifiable against external sources, not opinions or predictions.
Evidence retrieval
Up to 5 web sources are retrieved per claim using Tavily Search. If you specified trusted reference URLs when creating a job, those domains are prioritised in the search.
LLM evaluation
Claude evaluates the claim against the retrieved sources and produces a structured verdict: a status, a confidence score (0–1), a 2–3 sentence reasoning, and a per-source relevance assessment.
Credibility score
After all claims in a job are checked, a job-level credibility score is calculated from the distribution of verdicts and their confidence values.
Verdict types
Supported
Sources clearly and directly confirm the claim.
Partially supported
Sources confirm part of the claim or address it indirectly.
Refuted
Sources directly contradict the claim.
Inconclusive
No relevant sources found, or evidence is too weak to decide.
Confidence score
| Score | Meaning |
|---|---|
| 0.9 – 1.0 | Multiple independent sources agree and directly address the claim |
| 0.7 – 0.89 | One strong authoritative source directly addresses the claim |
| 0.5 – 0.69 | Sources are somewhat relevant but not definitive |
| < 0.5 | Evidence is weak, tangential, or contradictory |
Trusted / reference URLs
When creating a job you can supply one or more reference URLs (e.g. the original article, an official report, a press release). The domains of those URLs are prioritised in the evidence search for every claim in that job. This is useful when you want verdicts grounded in a specific authoritative source rather than general web results.
Fact-checking is capped at 20 claims per job across all plans. Claims beyond this cap are marked inconclusive automatically. You can manually dismiss claims you consider irrelevant — dismissed claims do not count toward the cap.
Processing Pipeline
Every audio file passes through 7 sequential stages. Each stage checkpoints its state so failed steps can retry from where they left off.
Ingestion
Accepts audio via URL or direct upload. Supports podcasts, interviews, debates, meetings, and any spoken-word media.
Transcription
Powered by AssemblyAI with speaker diarization — produces a full verbatim transcript annotated with speaker labels and word-level timestamps.
Segmentation
Splits the transcript into coherent topic segments. Each segment gets its own summary, topic label, and time range.
Analysis
Claude extracts topics, claims, and sentiment in parallel. Identifies factual assertions, opinions, and key discussion points.
Embedding
Transcript segments are embedded via OpenAI and stored in Qdrant, enabling semantic search and conversational Q&A.
Fact-Checking
Each extracted claim is independently verified: web search retrieves up to 5 sources per claim, then Claude evaluates the evidence and issues a verdict with a confidence score.
Report
Compiles a structured intelligence report with verdict summaries, topic breakdown, speaker analysis, and a shareable public URL.
Content Types
Audionyx adapts its analysis based on the content type you select.
| Type | Best for |
|---|---|
| Podcast | Interview-style shows, monologues, audio essays |
| Interview | Q&A sessions, press interviews, research conversations |
| Debate | Structured arguments, panel discussions with opposing views |
| Panel | Multi-speaker roundtables, conference sessions |
| Meeting | Internal calls, standups, client meetings |
| Tech Talk | Conference talks, technical presentations, lectures |
Plans & Limits
| Plan | Jobs / mo | Max duration |
|---|---|---|
Free $0 forever | — browse & view public reports only | — |
Pro $14 / month | 20 | 2 hours |
Newsroom $39 / seat / month | 100 shared pool | 4 hours |
A 7-day trial with Pro limits is available on new accounts when the trial feature is active.
API Reference
Coming SoonA public REST API is planned for Newsroom plan subscribers. It will expose job creation, transcript retrieval, claim queries, and webhooks for pipeline events. Documentation will be published here when the API enters beta.
View API Reference page