{% extends "base.html" %} {% block title %}Speech-to-text · {{ profile.name }}{% endblock %} {% block content %}

Speech-to-text benchmark · {{ profile.name }}

Benchmarks an STT model's real-time-factor (Lemonade's whispercpp/moonshine recipes, a multipart /v1/audio/transcriptions). Deliberately separate from the model/router leaderboard -- real-time-factor isn't comparable to LLM token throughput. This runs synchronously.

{% if capabilities.error %}

Couldn't reach this instance ({{ capabilities.error }}) -- engine and backend fall back to free text.

{% endif %}
{% if stt_models %} {% else %}
No STT models pulled yet on this profile -- pull one via lemonmatrix profile pull-model (recipe whispercpp), then reload this page.
{% endif %}
{% if stt_backends %} {% else %}
No STT backend detected -- install it on the profile page first, or enter it manually.
{% endif %}
{% if default_os %}
{{ default_os }} (fixed by this profile)
{% else %}
Couldn't detect -- see profile page to fix discovery.
{% endif %}
Must be a WAV file -- its exact duration (read from its header) is what real_time_factor is computed against.

STT results for {{ profile.name }}

{% if not results %}

No STT results yet.

{% else %}
{% for r in results %} {% endfor %}
TimestampModelTranscription (ms)Audio (s)RTFLanguageValidNotes
{{ r.timestamp }} {{ r.model.name }} {{ "%.1f"|format(r.metrics.transcription_time_ms) }} {{ "%.2f"|format(r.metrics.audio_duration_s) }} {{ "%.2f"|format(r.metrics.real_time_factor) }}x {{ r.metrics.language or "–" }} {{ "valid" if r.validity.valid else "invalid" }} {{ r.validity.notes or "" }}
{% endif %}
{% endblock %}