Troubleshoot a problem step by step

Start with the affected environment: browser, native app or hardware recorder. The device on which you read help may be different from the affected device. Note the exact error and checks already completed. Passwords, API keys and interview content are not needed for this description.

A recording has no sound

1. Say a sentence and watch the audio level meter. On the web, it is available both in the recorder lobby before recording and in the recorder during recording. The native app and IoT recorder also have a level meter: movement indicates that an audio signal is coming in. If the meter moves, go to step 4; otherwise, check the following points.
2. In the browser or app, check microphone permission. Browser website permissions are beside the address bar. Camera and microphone permissions are separate. The operating system must also allow the browser or native app to use the microphone.
3. Check the microphone in use and the device selection where available. A connected headset may change the selected microphone. Check any mute switch, then speak again and watch the meter.
4. Make and listen to a short new test recording. A moving meter confirms the current input signal, but does not yet confirm that a recording was saved or uploaded correctly.
5. If the test recording is silent, check playback mute and the selected output device. Try another recording known to have sound. If it is also silent, investigate playback first.
6. If the test remains silent or the meter does not move, check another available microphone and, for browser or app recordings, whether another application is using the device. Note the affected device, browser/OS where applicable, and results.

Granting microphone access later cannot add sound to a file recorded without it. Keep the original: apparent silence can also be a playback problem.

Upload or synchronization does not finish

1. Keep the original file or local recording. Do not uninstall the app or clear app/browser data while recordings are not backed up.
2. Check connectivity and the exact status: uploading, transfer failed, or uploaded and processing? Transfer, media processing and transcription are separate stages.
3. Restore a lost connection and inspect the existing upload status. Automatic resumption is not guaranteed for every client or upload.
4. Check the active tenant and destination project. An uploaded file may be in another tenant or outside your access.
5. Follow specific file-size, storage, credit or rate-limit messages. Repeated immediate uploads do not remove these limits.
6. If the client offers retry, use it deliberately. Before starting an additional upload, check whether a session already exists to avoid duplicates.

For unresolved problems, note the time, client version, failing stage and error message.

Signing in to the native app

1. Distinguish the native app from an installed web app/PWA; their sign-in methods differ.
2. Check whether browser sign-in works. If not, investigate activation, invitation and credentials first.
3. Browser Microsoft SSO does not imply native SSO support. The reviewed native app implements password sign-in, not Microsoft SSO or magic links. Browser instructions for “Forgot password?” are not a documented native sign-in path. Check the installed app version before assuming a button exists.
4. Note the app version, OS and exact error. Ask your administrator whether a password login is intended for your account; do not bypass an organization-wide SSO requirement.
5. After signing in, check the selected tenant.

Reinstallation is not a safe default when local recordings are not backed up.

Different content or fields from a colleague

For additional fields in the project description, the exact field names matter. Without them, no most likely cause is established; first ask which fields differ. Roles and modules are things to check, not an explanation for an unidentified field.

One verified difference is Automatic summary, shown to admins when editing an existing project. The full creation form shows name, description, AI context and color without additional role or module conditions. See Project fields.

1. Open the avatar menu at the top right and compare the active tenant.
2. Compare roles, access assignments and enabled modules with your administrator. Your role can differ between tenants.
3. If you see each other’s content, open Projects → affected project → Share in the top action area. It lists direct members, the owner, admin access and groups. Sessions in an accessible project are shared with people who have project access. Admins and managers with project access can add or remove people and groups; users and participants can view the list. When removing a grant, also check other paths through groups or the admin role. Product help does not read this actual membership list itself.
4. For differing project fields, first compare tenant and page. If both match, note the exact additional or missing field names. A role or module difference alone does not establish why that field differs.
5. Some functions require a module. Product help cannot enable modules or change access.

On small screens, navigation may be collapsed. Open the described page or menu if a highlighted target is not visible.

Storage, account limits, credits and rate limits

These are different constraints. Storage concerns files, account limits concern account setup or use, credits concern paid processing, and rate limits concern too many requests within a period. Read the exact message first. Ask your administrator about credits or access; wait for the indicated time after a rate limit. Product help cannot inspect your storage usage or payment status.

Hardware recorder and installation

Check the device variant and firmware first. M5Stack CoreS3 SE and Waveshare E-Paper require different firmware and instructions. The StoryVault installer links the device manuals and configuration. USB installation requires a supported desktop browser with Web Serial, such as Chrome or Edge, and a data cable. A charging-only cable is insufficient; this is not an iPhone flashing workflow.

Check the cable/USB connection, selected device variant and the manual’s Wi-Fi requirements in order. For uploads, check the local recording, connection and device assignment in the correct tenant. Never share the device API key in a screenshot. Back up local recordings before reflashing. Planned OTA or context functions must not be treated as released; the installed firmware and supported server functions matter.

Transcript quality, dialects and languages

Listen to a short affected passage in the original: is speech clear, overlapping, reverberant or noisy? Another model cannot reliably reconstruct missing speech. Missing or incorrect transcript passages do not necessarily mean the original audio is missing. If the speech is understandable in the recording, you can correct the transcript or consider retranscribing it. A poor result alone does not establish whether the cause lies with the provider or StoryVault. Then check the provider and detected/configured language. Language support does not guarantee recognition of every dialect, or that all provider options exist in StoryVault. Speaker labels do not establish a person’s identity.

Gladia and Mistral

Mixed languages: Gladia documents model-dependent language support and a separate code-switching configuration. StoryVault’s current integration requests automatic language detection but does not explicitly configure code switching. Do not assume a corresponding StoryVault setting or guaranteed recognition of every language switch. Sources checked on 10 October 2026: languages, code switching.

Mistral language selection: Mistral documents a fixed language option, but not together with timestamps. StoryVault requests segment timestamps without a fixed language, so this path uses automatic detection. Up to 100 words or short phrases can be provided as vocabulary hints; this is optimized for English and experimental for other languages, not a correctness guarantee. Source checked on 10 October 2026: Mistral transcription and FAQ.

Speaker count: Diarization separates voices within a recording and does not verify identities. Gladia’s speaker-count hints are not guaranteed limits. Check difficult passages and overlaps against the original. Source checked on 10 October 2026: Gladia speaker diarization.

For support, describe provider, language/dialect, approximate error position and error type; do not upload interview content to product help.

Export a transcript: speaker and timestamp only at speaker changes, like the coding view

There is currently no dedicated export with speaker and start time only at the beginning of each speaker change. The coding view groups adjacent segments for display. TXT has speaker information per segment but no timestamps. SRT/VTT have segment timestamps.

For reformatting in a tool approved for your data, use this instruction:

Combine only immediately adjacent segments by the same speaker into a paragraph. Put the speaker name and existing start time only at the paragraph’s beginning. Preserve all wording and order; do not summarize, correct or add content. Mark uncertain speakers or times instead of inventing them.

Use an export with timestamps if you need start times; TXT alone is insufficient. Check completeness. Product help explains this process but does not process interviews.

Coding, processes, API and knowledge

Distinguish codebooks, text selection, accepting AI suggestions and coding reports. Open suggestions are not accepted codings. Check session selections and filters when results are missing. For BPMN, distinguish sources, description, diagram and approval status; a draft is not a confirmed process. Actions also depend on modules and permissions.

API/MCP requires appropriate access and credentials. Share error codes and request descriptions without keys. RAG uses selected, processed sources for answers; a knowledge graph shows extracted relationships that need review. These content features are separate from product help.

See coding, processes, API, knowledge stores and knowledge graph.