Caller on a bad signal — AI will switch engine mid-call to keep up.
Rural sites, underground car parks, patchy mobile coverage. When the audio degrades, a single-engine speech-to-text system produces garbled transcripts or fails silently. IBIOS will monitor transcription confidence in real time and switch to a fallback engine as soon as quality drops, maintaining transcript quality without the caller or staff noticing any disruption. The IBIOS Receptionist, including live-call STT fallback, is coming soon and is not live yet — this scenario shows how it will work.
Why is a single speech-to-text engine a single point of failure?
Many AI receptionist products rely on a single speech-to-text engine. When audio quality is good, results are excellent. When audio degrades — a caller driving through a rural area, calling from a basement, or on a low-quality mobile connection — the engine starts producing errors. Words are misheard, names are mangled, numbers are wrong.
The AI continues the conversation based on what it thinks it heard. If it thinks the customer said a different address or a different product, it may update the wrong record or confirm the wrong information. The transcript that arrives in the CRM contains errors that are not labelled as errors. Staff act on them as if they were correct.
For a trade business whose callers often ring from sites rather than offices, this is not a rare edge case. It is a regular situation that requires a reliable response.
How will the AI receptionist handle poor-signal calls (coming soon)?
From first ring to completed transcript, even when the signal is unreliable.
Caller phones from a poor-signal location
A customer or installer calls from a rural site. Their mobile signal is intermittent. The audio will arrive at the IBIOS processing layer (coming soon) with gaps, compression artefacts, and variable bitrate.
Primary STT engine processes the call
The primary speech-to-text engine will begin transcribing. For the first part of the call where signal is acceptable, it performs normally. As the signal degrades, confidence scores on individual words begin to fall.
Confidence drop is detected
The monitoring layer will identify that rolling confidence has fallen below the threshold. As soon as it detects degradation, it will initiate a handoff to the first fallback engine in the chain.
Fallback engine takes over instantly
The fallback engine will pick up the audio stream mid-call. The AI Receptionist (coming soon) will continue the conversation without any interruption, reset, or audible change.
Transcript quality is maintained
The aim is to keep the transcript usable even on degraded audio. The completed transcript in IBIOS will mark fallback segments with their confidence indicators. Staff can see exactly which parts of the call were affected and act accordingly.
What will change for the business
The result is a system that performs reliably for callers who are most likely to be important to your business: installers calling from site, customers in rural areas, trade contacts on the road. These are not fringe cases. They are the everyday reality of a trade business, and they deserve the same transcript quality as a call made from a landline in a quiet office.
Questions about STT fallback and audio quality
Which STT engines does IBIOS use and in what order?
Coming soon. Live-call STT fallback is part of the AI Receptionist, which is not live yet. Once it launches, IBIOS will use multiple independent speech engines with automatic switching. The primary engine will be selected for low latency and accuracy under normal conditions. The first backup engine will be optimised for fragmented or low-bitrate audio where the primary engine loses confidence. A further engine will be a robust general-purpose model for highly degraded audio, including heavy compression artefacts. The specific engines used can be disclosed on request under NDA.
How does the system detect that the primary engine is struggling?
Once live (coming soon), the system will monitor the per-word confidence scores returned by the primary STT engine on a rolling basis. When the average confidence across a short window drops below the configured threshold, the orchestration layer will flag the session as degraded and trigger a handoff to the next engine in the chain. The threshold is designed to switch before transcript quality degrades to the point of unreliability, not after.
Does the caller hear any disruption when the engine switches?
No. Once the AI Receptionist launches (coming soon), the engine switch will happen at the transcription layer, not the audio layer. The caller’s audio will continue to flow uninterrupted, and the AI Receptionist will continue the conversation without any pause or reset. The only difference will be that the transcription engine processing their words has changed behind the scenes.
Is the transcript marked to indicate which engine produced which segment?
Yes, once live-call STT fallback launches (coming soon). In the IBIOS transcript viewer, segments produced by the fallback engine will be marked with a degraded-audio indicator alongside the confidence score for that segment. This allows staff reviewing the transcript to identify which parts of the conversation were transcribed under difficult conditions and to treat those segments with appropriate caution. Low-confidence segments will be highlighted rather than silently included at face value.
Can this also handle heavy accents or non-standard English?
The multi-engine system (coming soon) is designed primarily for audio quality degradation, but the confidence-monitoring approach will also benefit callers with heavy accents or non-standard speech patterns. If the primary engine consistently returns low confidence for a particular caller’s voice, it will trigger automatic engine switching in the same way as a poor-signal call. The backup engines have broader training on accent variation. For businesses in regions with distinctive local accents, this is expected to improve transcript accuracy across the board, not just on bad-signal calls.
Bad signal should not mean a bad transcript.
Coming soon: IBIOS will keep transcription reliable no matter where your callers are calling from. Book a walkthrough to hear how the fallback system will work.