{
  "_published": {
    "source": "docs/measurements/whisper-transcription-2026-11-08.json",
    "published": "2026-11-08",
    "publisher": "J4SGON S.L., vorluxai.com",
    "index": "https://vorluxai.com/data/",
    "cleaned": "Copied from our internal repository by scripts/publish_measurements.py. Every number is unchanged. Redacted: home directories -> ~ (1). Hostnames (spark-43d5, spark-7180) and the PRIME / spark2 labels are kept: the posts use them.",
    "redactions": {
      "home_path": 1
    }
  },
  "measured_at": "2026-11-08T08:02:25+00:00",
  "load_average_at_end": [
    4.65,
    4.33,
    3.16
  ],
  "finished_at": "2026-11-08T08:14:14+00:00",
  "host": "spark-43d5",
  "machine": "NVIDIA DGX Spark, GB10, 20-core Arm CPU, 128 GB unified memory (PRIME, spark-43d5)",
  "kernel": "6.17.0-1032-nvidia",
  "gpu_driver": "NVIDIA GB10, 580.173.02",
  "load_average_at_start": [
    1.42,
    1.35,
    1.6
  ],
  "gpu_processes_at_start": [
    "2998, ~/dev/ComfyUI/.venv/bin/python, 6977 MiB",
    "6752, /usr/libexec/gnome-remote-desktop-daemon, 176 MiB",
    "17940, /usr/local/lib/ollama/llama-server, 890 MiB",
    "14191, /app/.venv/bin/python3, 1240 MiB",
    "20788, /usr/local/lib/ollama/llama-server, 543 MiB",
    "3062865, VLLM::EngineCore, 1876 MiB",
    "3189762, /usr/local/lib/ollama/llama-server, 507 MiB",
    "3407366, /usr/local/lib/ollama/llama-server, 870 MiB",
    "1435472, /usr/local/lib/ollama/llama-server, 8993 MiB"
  ],
  "concurrent_workloads_note": "ComfyUI process resident on PRIME's GPU at start (see gpu_processes_at_start); the coordinator reported the news pipeline running on the second box (spark-7180), which this script did not check; Whisper itself ran on the CPU",
  "stt": "ghcr.io/speaches-ai/speaches:latest-cuda sha256:5b3bf06e65be312c140cf45823a9a17037206cddbf7dab3332e0ba6c05b63834",
  "stt_runtime": {
    "probe": "ctranslate2 4.5.0 faster_whisper 1.1.1 cuda_devices 0 cpu_compute_types ['float32', 'int8', 'int8_float32']",
    "cuda_compute_types": "Traceback (most recent call last):\n  File \"<string>\", line 1, in <module>\nValueError: This CTranslate2 package was not compiled with CUDA support",
    "speaches_whisper_config": "inference_device: Device = \"auto\"\n    compute_type: Quantization = \"default\"  # TODO: should this even be a configuration option?\n    cpu_threads: int = 0\n    ttl: int = Field(default=300, ge=-1)",
    "whisper_env_overrides": "none (no WHISPER__* variables in the container environment)",
    "ctranslate2_log": "[2026-10-24 12:15:18.611] [ctranslate2] [thread 78544] [warning] The compute type inferred from the saved model is float16, but the target device or backend do not support efficient float16 computation. The model weights have been automatically converted to use the float32 compute type instead."
  },
  "command": "python3 scripts/bench/whisper_transcription.py --reps 3 --clip-dir <tmp> --note '<concurrent_workloads_note>' --out docs/measurements/whisper-transcription-2026-11-08.json",
  "method": "two synthetic meeting clips (en: Piper, es: Kokoro); per model and clip 1 warm-up request (loads the model, discarded) + N timed requests to /v1/audio/transcriptions, language set per clip, temperature 0; processing time = client wall seconds; RTF = processing / audio seconds; WER vs the script after lower-casing and stripping punctuation, accents kept (wer) and stripped (wer_accents_stripped); numbers written as digits count as errors; synthetic TTS speech is cleaner than a real room, so WER here is a floor; memory = docker stats of the Speaches container (cgroup memory) before the model's warm-up and after its runs",
  "clips": {
    "en": {
      "script": "Okay, let's start. Three things today. First, the quarterly close. Javier says eleven clients still owe us their September invoices, so Pablo will call them on Tuesday and Wednesday. Second, the new hires. Lucia has filed one social security registration and is waiting for an identity document for the other one. Third, the AI assistant. It saves time on simple replies, but it gets amounts wrong, so from now on nobody sends an email with an amount in it without a human check. Next meeting is Monday the nineteenth at nine.",
      "script_words": 92,
      "tts": "openedai-speech (container j4sgon-speech-tts-1) model tts-1 (Piper) voice alloy speed 1.0",
      "audio_seconds": 24.49,
      "bytes": 1079886,
      "sha256": "2f4dcb6a7d006546d1aa3c8132d3b23014040a586a024e4f48df2066fa5665f4",
      "matches_2026_10_17_clip": false
    },
    "es": {
      "script": "Bueno, empezamos. Hoy tenemos tres temas. Primero, el cierre del trimestre. Javier dice que once clientes todavía nos deben las facturas de septiembre, así que Pablo los llamará el martes y el miércoles. Segundo, las nuevas contrataciones. Lucía ha tramitado un alta en la Seguridad Social y está esperando el documento de identidad de la otra persona. Tercero, el asistente de inteligencia artificial. Ahorra tiempo en las respuestas sencillas, pero se equivoca con los importes, así que a partir de ahora nadie envía un correo con un importe sin que lo revise una persona. La próxima reunión es el lunes diecinueve a las nueve.",
      "script_words": 104,
      "tts": "Kokoro-FastAPI (container j4sgon-speech-kokoro-1) model kokoro voice ef_dora lang_code e speed 1.0",
      "audio_seconds": 36.01,
      "bytes": 1728644,
      "sha256": "565fe5d43a9dfdf9278daecf23e6ed8821063e8720e64ed4b437eaf93ecd5de5"
    }
  },
  "results": [
    {
      "model": "Systran/faster-whisper-tiny.en",
      "clip": "en",
      "median_processing_s": 1.14,
      "median_rtf": 0.046,
      "words": 91,
      "wer": 0.065,
      "wer_accents_stripped": 0.065,
      "runs": [
        {
          "run": 1,
          "processing_s": 1.137,
          "rtf": 0.046,
          "words": 91,
          "wer": 0.065,
          "wer_accents_stripped": 0.065,
          "text": "Okay, let's start. Three things today. First, the quarterly close. Javier says 11 clients still owe us their September invoices, so Pablo will call them on Tuesday and Wednesday. Second, the new hires. Lucia has filed one social security registration and is waiting for an identity document for the other one. Third, the eye assistant saves time on simple replies, but it gets amounts wrong. So from now on, nobody sends an email with an amount in it without a human check. X meeting is Monday, the 19th at 9."
        },
        {
          "run": 2,
          "processing_s": 1.123,
          "rtf": 0.046,
          "words": 91,
          "wer": 0.065,
          "wer_accents_stripped": 0.065,
          "text": "Okay, let's start. Three things today. First, the quarterly close. Javier says 11 clients still owe us their September invoices, so Pablo will call them on Tuesday and Wednesday. Second, the new hires. Lucia has filed one social security registration and is waiting for an identity document for the other one. Third, the eye assistant saves time on simple replies, but it gets amounts wrong. So from now on, nobody sends an email with an amount in it without a human check. X meeting is Monday, the 19th at 9."
        },
        {
          "run": 3,
          "processing_s": 1.169,
          "rtf": 0.048,
          "words": 91,
          "wer": 0.065,
          "wer_accents_stripped": 0.065,
          "text": "Okay, let's start. Three things today. First, the quarterly close. Javier says 11 clients still owe us their September invoices, so Pablo will call them on Tuesday and Wednesday. Second, the new hires. Lucia has filed one social security registration and is waiting for an identity document for the other one. Third, the eye assistant saves time on simple replies, but it gets amounts wrong. So from now on, nobody sends an email with an amount in it without a human check. X meeting is Monday, the 19th at 9."
        }
      ],
      "host_load_average_before": [
        1.42,
        1.35,
        1.6
      ],
      "host_load_average_after": [
        1.39,
        1.35,
        1.59
      ],
      "waited_for_load_s": 0,
      "speaches_loaded_models_after_runs": [
        "Systran/faster-whisper-small",
        "Systran/faster-whisper-tiny.en"
      ],
      "speaches_container_mem_before_load": "1.479GiB / 121.7GiB",
      "speaches_container_mem_after_runs": "1.677GiB / 121.7GiB",
      "unload": "not requested; left to Speaches' idle TTL (DELETE /api/ps deadlocks this build)"
    },
    {
      "model": "Systran/faster-whisper-base.en",
      "clip": "en",
      "median_processing_s": 2.55,
      "median_rtf": 0.104,
      "words": 92,
      "wer": 0.043,
      "wer_accents_stripped": 0.043,
      "runs": [
        {
          "run": 1,
          "processing_s": 2.534,
          "rtf": 0.103,
          "words": 92,
          "wer": 0.043,
          "wer_accents_stripped": 0.043,
          "text": "Okay, let's start. Three things today. First, the quarterly close. Javier says 11 clients still owe us their September invoices, so Pablo will call them on Tuesday and Wednesday. Second, the new hires. Lucia has filed one Social Security registration and is waiting for an identity document for the other one. Third, the I-assistant. It saves time on simple replies, but it gets amounts wrong, so from now on nobody sends an email with an amount in it without a human check. Next meeting is Monday, the 19th at 9."
        },
        {
          "run": 2,
          "processing_s": 2.659,
          "rtf": 0.109,
          "words": 92,
          "wer": 0.043,
          "wer_accents_stripped": 0.043,
          "text": "Okay, let's start. Three things today. First, the quarterly close. Javier says 11 clients still owe us their September invoices, so Pablo will call them on Tuesday and Wednesday. Second, the new hires. Lucia has filed one Social Security registration and is waiting for an identity document for the other one. Third, the I-assistant. It saves time on simple replies, but it gets amounts wrong, so from now on nobody sends an email with an amount in it without a human check. Next meeting is Monday, the 19th at 9."
        },
        {
          "run": 3,
          "processing_s": 2.552,
          "rtf": 0.104,
          "words": 92,
          "wer": 0.043,
          "wer_accents_stripped": 0.043,
          "text": "Okay, let's start. Three things today. First, the quarterly close. Javier says 11 clients still owe us their September invoices, so Pablo will call them on Tuesday and Wednesday. Second, the new hires. Lucia has filed one Social Security registration and is waiting for an identity document for the other one. Third, the I-assistant. It saves time on simple replies, but it gets amounts wrong, so from now on nobody sends an email with an amount in it without a human check. Next meeting is Monday, the 19th at 9."
        }
      ],
      "host_load_average_before": [
        1.44,
        1.36,
        1.6
      ],
      "host_load_average_after": [
        1.83,
        1.45,
        1.62
      ],
      "waited_for_load_s": 0,
      "speaches_loaded_models_after_runs": [
        "Systran/faster-whisper-small",
        "Systran/faster-whisper-tiny.en",
        "Systran/faster-whisper-base.en"
      ],
      "speaches_container_mem_before_load": "1.677GiB / 121.7GiB",
      "speaches_container_mem_after_runs": "2.003GiB / 121.7GiB",
      "unload": "not requested; left to Speaches' idle TTL (DELETE /api/ps deadlocks this build)"
    },
    {
      "model": "Systran/faster-whisper-small",
      "clip": "en",
      "median_processing_s": 7.0,
      "median_rtf": 0.286,
      "words": 91,
      "wer": 0.054,
      "wer_accents_stripped": 0.054,
      "runs": [
        {
          "run": 1,
          "processing_s": 7.163,
          "rtf": 0.293,
          "words": 91,
          "wer": 0.054,
          "wer_accents_stripped": 0.054,
          "text": "Okay, let's start. Three things today. First, the quarterly close. Javier says 11 clients still owe us their September invoices, so Pablo will call them on Tuesday and Wednesday. Second, the new hires. Lucia has filed one Social Security registration and is waiting for an identity document for the other one. Third, the eye assistant saves time on simple replies, but it gets amounts wrong, so from now on nobody sends an email with an amount in it without a human check. Next meeting is Monday, the 19th at 9."
        },
        {
          "run": 2,
          "processing_s": 6.997,
          "rtf": 0.286,
          "words": 91,
          "wer": 0.054,
          "wer_accents_stripped": 0.054,
          "text": "Okay, let's start. Three things today. First, the quarterly close. Javier says 11 clients still owe us their September invoices, so Pablo will call them on Tuesday and Wednesday. Second, the new hires. Lucia has filed one Social Security registration and is waiting for an identity document for the other one. Third, the eye assistant saves time on simple replies, but it gets amounts wrong, so from now on nobody sends an email with an amount in it without a human check. Next meeting is Monday, the 19th at 9."
        },
        {
          "run": 3,
          "processing_s": 6.87,
          "rtf": 0.281,
          "words": 91,
          "wer": 0.054,
          "wer_accents_stripped": 0.054,
          "text": "Okay, let's start. Three things today. First, the quarterly close. Javier says 11 clients still owe us their September invoices, so Pablo will call them on Tuesday and Wednesday. Second, the new hires. Lucia has filed one Social Security registration and is waiting for an identity document for the other one. Third, the eye assistant saves time on simple replies, but it gets amounts wrong, so from now on nobody sends an email with an amount in it without a human check. Next meeting is Monday, the 19th at 9."
        }
      ],
      "host_load_average_before": [
        1.84,
        1.46,
        1.63
      ],
      "host_load_average_after": [
        3.74,
        2.06,
        1.82
      ],
      "waited_for_load_s": 0,
      "speaches_loaded_models_after_runs": [
        "Systran/faster-whisper-small",
        "Systran/faster-whisper-tiny.en",
        "Systran/faster-whisper-base.en"
      ],
      "speaches_container_mem_before_load": "2.003GiB / 121.7GiB",
      "speaches_container_mem_after_runs": "2.076GiB / 121.7GiB",
      "unload": "not requested; left to Speaches' idle TTL (DELETE /api/ps deadlocks this build)"
    },
    {
      "model": "Systran/faster-whisper-small",
      "clip": "es",
      "median_processing_s": 8.11,
      "median_rtf": 0.225,
      "words": 104,
      "wer": 0.048,
      "wer_accents_stripped": 0.048,
      "runs": [
        {
          "run": 1,
          "processing_s": 8.214,
          "rtf": 0.228,
          "words": 104,
          "wer": 0.048,
          "wer_accents_stripped": 0.048,
          "text": "Bueno, empezamos. Hoy tenemos tres temas. Primero, el cierre del trimestre. Javier dice que 11 clientes todavía nos deben las facturas de septiembre, así que Pablo los llamará el martes y el miércoles. Segundo, las nuevas contrataciones. Lucía ha tramitado un alta en la seguridad social y está esperando el documento de identidad de la otra persona. Tercero, el asistente de inteligencia artificial. Ahora tiempo en las respuestas sencillas, no se equivoca con los importes, así que a partir de ahora nadie envía un correo con un importe sin que lo revise una persona. La próxima reunión es el lunes 19 a las 9."
        },
        {
          "run": 2,
          "processing_s": 8.075,
          "rtf": 0.224,
          "words": 104,
          "wer": 0.048,
          "wer_accents_stripped": 0.048,
          "text": "Bueno, empezamos. Hoy tenemos tres temas. Primero, el cierre del trimestre. Javier dice que 11 clientes todavía nos deben las facturas de septiembre, así que Pablo los llamará el martes y el miércoles. Segundo, las nuevas contrataciones. Lucía ha tramitado un alta en la seguridad social y está esperando el documento de identidad de la otra persona. Tercero, el asistente de inteligencia artificial. Ahora tiempo en las respuestas sencillas, no se equivoca con los importes, así que a partir de ahora nadie envía un correo con un importe sin que lo revise una persona. La próxima reunión es el lunes 19 a las 9."
        },
        {
          "run": 3,
          "processing_s": 8.106,
          "rtf": 0.225,
          "words": 104,
          "wer": 0.048,
          "wer_accents_stripped": 0.048,
          "text": "Bueno, empezamos. Hoy tenemos tres temas. Primero, el cierre del trimestre. Javier dice que 11 clientes todavía nos deben las facturas de septiembre, así que Pablo los llamará el martes y el miércoles. Segundo, las nuevas contrataciones. Lucía ha tramitado un alta en la seguridad social y está esperando el documento de identidad de la otra persona. Tercero, el asistente de inteligencia artificial. Ahora tiempo en las respuestas sencillas, no se equivoca con los importes, así que a partir de ahora nadie envía un correo con un importe sin que lo revise una persona. La próxima reunión es el lunes 19 a las 9."
        }
      ],
      "host_load_average_before": [
        1.84,
        1.46,
        1.63
      ],
      "host_load_average_after": [
        3.74,
        2.06,
        1.82
      ],
      "waited_for_load_s": 0,
      "speaches_loaded_models_after_runs": [
        "Systran/faster-whisper-small",
        "Systran/faster-whisper-tiny.en",
        "Systran/faster-whisper-base.en"
      ],
      "speaches_container_mem_before_load": "2.003GiB / 121.7GiB",
      "speaches_container_mem_after_runs": "2.076GiB / 121.7GiB",
      "unload": "not requested; left to Speaches' idle TTL (DELETE /api/ps deadlocks this build)"
    },
    {
      "model": "Systran/faster-whisper-medium",
      "clip": "en",
      "median_processing_s": 20.47,
      "median_rtf": 0.836,
      "words": 92,
      "wer": 0.043,
      "wer_accents_stripped": 0.043,
      "runs": [
        {
          "run": 1,
          "processing_s": 20.187,
          "rtf": 0.824,
          "words": 92,
          "wer": 0.043,
          "wer_accents_stripped": 0.043,
          "text": "Okay, let's start. Three things today. First, the quarterly close. Javier says 11 clients still owe us their September invoices, so Pablo will call them on Tuesday and Wednesday. Second, the new hires. Lucia has filed one Social Security registration and is waiting for an identity document for the other one. Third, the eye assistant. It saves time on simple replies, but it gets amounts wrong, so from now on nobody sends an email with an amount in it without a human check. Next meeting is Monday the 19th at 9."
        },
        {
          "run": 2,
          "processing_s": 20.614,
          "rtf": 0.842,
          "words": 92,
          "wer": 0.043,
          "wer_accents_stripped": 0.043,
          "text": "Okay, let's start. Three things today. First, the quarterly close. Javier says 11 clients still owe us their September invoices, so Pablo will call them on Tuesday and Wednesday. Second, the new hires. Lucia has filed one Social Security registration and is waiting for an identity document for the other one. Third, the eye assistant. It saves time on simple replies, but it gets amounts wrong, so from now on nobody sends an email with an amount in it without a human check. Next meeting is Monday the 19th at 9."
        },
        {
          "run": 3,
          "processing_s": 20.474,
          "rtf": 0.836,
          "words": 92,
          "wer": 0.043,
          "wer_accents_stripped": 0.043,
          "text": "Okay, let's start. Three things today. First, the quarterly close. Javier says 11 clients still owe us their September invoices, so Pablo will call them on Tuesday and Wednesday. Second, the new hires. Lucia has filed one Social Security registration and is waiting for an identity document for the other one. Third, the eye assistant. It saves time on simple replies, but it gets amounts wrong, so from now on nobody sends an email with an amount in it without a human check. Next meeting is Monday the 19th at 9."
        }
      ],
      "host_load_average_before": [
        3.74,
        2.06,
        1.82
      ],
      "host_load_average_after": [
        4.48,
        3.09,
        2.26
      ],
      "waited_for_load_s": 0,
      "speaches_loaded_models_after_runs": [
        "Systran/faster-whisper-small",
        "Systran/faster-whisper-tiny.en",
        "Systran/faster-whisper-base.en",
        "Systran/faster-whisper-medium"
      ],
      "speaches_container_mem_before_load": "2.076GiB / 121.7GiB",
      "speaches_container_mem_after_runs": "6.436GiB / 121.7GiB",
      "unload": "not requested; left to Speaches' idle TTL (DELETE /api/ps deadlocks this build)"
    },
    {
      "model": "Systran/faster-whisper-medium",
      "clip": "es",
      "median_processing_s": 23.12,
      "median_rtf": 0.642,
      "words": 104,
      "wer": 0.029,
      "wer_accents_stripped": 0.029,
      "runs": [
        {
          "run": 1,
          "processing_s": 22.724,
          "rtf": 0.631,
          "words": 104,
          "wer": 0.029,
          "wer_accents_stripped": 0.029,
          "text": "Bueno, empezamos. Hoy tenemos tres temas. Primero, el cierre del trimestre. Javier dice que once clientes todavía nos deben las facturas de septiembre, así que Pablo los llamará el martes y el miércoles. Segundo, las nuevas contrataciones. Lucía ha tramitado un alta en la seguridad social y está esperando el documento de identidad de la otra persona. Tercero, el asistente de inteligencia artificial. Ahora tiempo en las respuestas sencillas, pero se equivoca con los importes, así que a partir de ahora nadie envía un correo con un importe sin que lo revise una persona. La próxima reunión es el lunes 19 a las 9."
        },
        {
          "run": 2,
          "processing_s": 23.122,
          "rtf": 0.642,
          "words": 104,
          "wer": 0.029,
          "wer_accents_stripped": 0.029,
          "text": "Bueno, empezamos. Hoy tenemos tres temas. Primero, el cierre del trimestre. Javier dice que once clientes todavía nos deben las facturas de septiembre, así que Pablo los llamará el martes y el miércoles. Segundo, las nuevas contrataciones. Lucía ha tramitado un alta en la seguridad social y está esperando el documento de identidad de la otra persona. Tercero, el asistente de inteligencia artificial. Ahora tiempo en las respuestas sencillas, pero se equivoca con los importes, así que a partir de ahora nadie envía un correo con un importe sin que lo revise una persona. La próxima reunión es el lunes 19 a las 9."
        },
        {
          "run": 3,
          "processing_s": 23.296,
          "rtf": 0.647,
          "words": 104,
          "wer": 0.029,
          "wer_accents_stripped": 0.029,
          "text": "Bueno, empezamos. Hoy tenemos tres temas. Primero, el cierre del trimestre. Javier dice que once clientes todavía nos deben las facturas de septiembre, así que Pablo los llamará el martes y el miércoles. Segundo, las nuevas contrataciones. Lucía ha tramitado un alta en la seguridad social y está esperando el documento de identidad de la otra persona. Tercero, el asistente de inteligencia artificial. Ahora tiempo en las respuestas sencillas, pero se equivoca con los importes, así que a partir de ahora nadie envía un correo con un importe sin que lo revise una persona. La próxima reunión es el lunes 19 a las 9."
        }
      ],
      "host_load_average_before": [
        3.74,
        2.06,
        1.82
      ],
      "host_load_average_after": [
        4.48,
        3.09,
        2.26
      ],
      "waited_for_load_s": 0,
      "speaches_loaded_models_after_runs": [
        "Systran/faster-whisper-small",
        "Systran/faster-whisper-tiny.en",
        "Systran/faster-whisper-base.en",
        "Systran/faster-whisper-medium"
      ],
      "speaches_container_mem_before_load": "2.076GiB / 121.7GiB",
      "speaches_container_mem_after_runs": "6.436GiB / 121.7GiB",
      "unload": "not requested; left to Speaches' idle TTL (DELETE /api/ps deadlocks this build)"
    },
    {
      "model": "deepdml/faster-whisper-large-v3-turbo-ct2",
      "clip": "en",
      "median_processing_s": 12.18,
      "median_rtf": 0.497,
      "words": 92,
      "wer": 0.043,
      "wer_accents_stripped": 0.043,
      "runs": [
        {
          "run": 1,
          "processing_s": 12.18,
          "rtf": 0.497,
          "words": 92,
          "wer": 0.043,
          "wer_accents_stripped": 0.043,
          "text": "Okay let's start. Three things today. First, the quarterly close. Javier says 11 clients still owe us their September invoices, so Pablo will call them on Tuesday and Wednesday. Second, the new hires. Lucia has filed one social security registration and is waiting for an identity document for the other one. Third, the eye assistant. It saves time on simple replies, but it gets amounts wrong, so from now on nobody sends an email with an amount in it without a human check. Next meeting is Monday the 19th at 9."
        },
        {
          "run": 2,
          "processing_s": 11.859,
          "rtf": 0.484,
          "words": 92,
          "wer": 0.043,
          "wer_accents_stripped": 0.043,
          "text": "Okay let's start. Three things today. First, the quarterly close. Javier says 11 clients still owe us their September invoices, so Pablo will call them on Tuesday and Wednesday. Second, the new hires. Lucia has filed one social security registration and is waiting for an identity document for the other one. Third, the eye assistant. It saves time on simple replies, but it gets amounts wrong, so from now on nobody sends an email with an amount in it without a human check. Next meeting is Monday the 19th at 9."
        },
        {
          "run": 3,
          "processing_s": 12.56,
          "rtf": 0.513,
          "words": 92,
          "wer": 0.043,
          "wer_accents_stripped": 0.043,
          "text": "Okay let's start. Three things today. First, the quarterly close. Javier says 11 clients still owe us their September invoices, so Pablo will call them on Tuesday and Wednesday. Second, the new hires. Lucia has filed one social security registration and is waiting for an identity document for the other one. Third, the eye assistant. It saves time on simple replies, but it gets amounts wrong, so from now on nobody sends an email with an amount in it without a human check. Next meeting is Monday the 19th at 9."
        }
      ],
      "host_load_average_before": [
        4.48,
        3.09,
        2.26
      ],
      "host_load_average_after": [
        3.94,
        3.38,
        2.49
      ],
      "waited_for_load_s": 0,
      "speaches_loaded_models_after_runs": [
        "Systran/faster-whisper-medium",
        "deepdml/faster-whisper-large-v3-turbo-ct2"
      ],
      "speaches_container_mem_before_load": "6.436GiB / 121.7GiB",
      "speaches_container_mem_after_runs": "8.847GiB / 121.7GiB",
      "unload": "not requested; left to Speaches' idle TTL (DELETE /api/ps deadlocks this build)"
    },
    {
      "model": "deepdml/faster-whisper-large-v3-turbo-ct2",
      "clip": "es",
      "median_processing_s": 22.22,
      "median_rtf": 0.617,
      "words": 104,
      "wer": 0.038,
      "wer_accents_stripped": 0.038,
      "runs": [
        {
          "run": 1,
          "processing_s": 21.653,
          "rtf": 0.601,
          "words": 104,
          "wer": 0.038,
          "wer_accents_stripped": 0.038,
          "text": "Bueno, empezamos. Hoy tenemos tres temas. Primero, el cierre del trimestre. Javier dice que 11 clientes todavía nos deben las facturas de septiembre, así que Pablo los llamará el martes y el miércoles. Segundo, las nuevas contrataciones. Lucía ha tramitado un alta en la seguridad social y está esperando el documento de identidad de la otra persona. Tercero, el asistente de inteligencia artificial. Ahora tiempo en las respuestas sencillas, pero se equivoca con los importes, así que a partir de ahora nadie envía un correo con un importe sin que lo revise una persona. La próxima reunión es el lunes 19 a las 9."
        },
        {
          "run": 2,
          "processing_s": 22.442,
          "rtf": 0.623,
          "words": 104,
          "wer": 0.038,
          "wer_accents_stripped": 0.038,
          "text": "Bueno, empezamos. Hoy tenemos tres temas. Primero, el cierre del trimestre. Javier dice que 11 clientes todavía nos deben las facturas de septiembre, así que Pablo los llamará el martes y el miércoles. Segundo, las nuevas contrataciones. Lucía ha tramitado un alta en la seguridad social y está esperando el documento de identidad de la otra persona. Tercero, el asistente de inteligencia artificial. Ahora tiempo en las respuestas sencillas, pero se equivoca con los importes, así que a partir de ahora nadie envía un correo con un importe sin que lo revise una persona. La próxima reunión es el lunes 19 a las 9."
        },
        {
          "run": 3,
          "processing_s": 22.222,
          "rtf": 0.617,
          "words": 104,
          "wer": 0.038,
          "wer_accents_stripped": 0.038,
          "text": "Bueno, empezamos. Hoy tenemos tres temas. Primero, el cierre del trimestre. Javier dice que 11 clientes todavía nos deben las facturas de septiembre, así que Pablo los llamará el martes y el miércoles. Segundo, las nuevas contrataciones. Lucía ha tramitado un alta en la seguridad social y está esperando el documento de identidad de la otra persona. Tercero, el asistente de inteligencia artificial. Ahora tiempo en las respuestas sencillas, pero se equivoca con los importes, así que a partir de ahora nadie envía un correo con un importe sin que lo revise una persona. La próxima reunión es el lunes 19 a las 9."
        }
      ],
      "host_load_average_before": [
        4.48,
        3.09,
        2.26
      ],
      "host_load_average_after": [
        3.94,
        3.38,
        2.49
      ],
      "waited_for_load_s": 0,
      "speaches_loaded_models_after_runs": [
        "Systran/faster-whisper-medium",
        "deepdml/faster-whisper-large-v3-turbo-ct2"
      ],
      "speaches_container_mem_before_load": "6.436GiB / 121.7GiB",
      "speaches_container_mem_after_runs": "8.847GiB / 121.7GiB",
      "unload": "not requested; left to Speaches' idle TTL (DELETE /api/ps deadlocks this build)"
    },
    {
      "model": "Systran/faster-whisper-large-v3",
      "clip": "en",
      "median_processing_s": 27.57,
      "median_rtf": 1.126,
      "words": 91,
      "wer": 0.054,
      "wer_accents_stripped": 0.054,
      "runs": [
        {
          "run": 1,
          "processing_s": 27.026,
          "rtf": 1.104,
          "words": 91,
          "wer": 0.054,
          "wer_accents_stripped": 0.054,
          "text": "Okay, let's start. Three things today. First, the quarterly close. Javier says 11 clients still owe us their September invoices, so Pablo will call them on Tuesday and Wednesday. Second, the new hires. Lucia has filed one social security registration and is waiting for an identity document for the other one. Third, the eye assistant. Saves time on simple replies, but it gets amounts wrong, so from now on nobody sends an email with an amount in it without a human check. Next meeting is Monday the 19th at 9."
        },
        {
          "run": 2,
          "processing_s": 27.909,
          "rtf": 1.14,
          "words": 91,
          "wer": 0.054,
          "wer_accents_stripped": 0.054,
          "text": "Okay, let's start. Three things today. First, the quarterly close. Javier says 11 clients still owe us their September invoices, so Pablo will call them on Tuesday and Wednesday. Second, the new hires. Lucia has filed one social security registration and is waiting for an identity document for the other one. Third, the eye assistant. Saves time on simple replies, but it gets amounts wrong, so from now on nobody sends an email with an amount in it without a human check. Next meeting is Monday the 19th at 9."
        },
        {
          "run": 3,
          "processing_s": 27.57,
          "rtf": 1.126,
          "words": 91,
          "wer": 0.054,
          "wer_accents_stripped": 0.054,
          "text": "Okay, let's start. Three things today. First, the quarterly close. Javier says 11 clients still owe us their September invoices, so Pablo will call them on Tuesday and Wednesday. Second, the new hires. Lucia has filed one social security registration and is waiting for an identity document for the other one. Third, the eye assistant. Saves time on simple replies, but it gets amounts wrong, so from now on nobody sends an email with an amount in it without a human check. Next meeting is Monday the 19th at 9."
        }
      ],
      "host_load_average_before": [
        3.62,
        3.32,
        2.48
      ],
      "host_load_average_after": [
        4.65,
        4.33,
        3.16
      ],
      "waited_for_load_s": 0,
      "speaches_loaded_models_after_runs": [
        "deepdml/faster-whisper-large-v3-turbo-ct2",
        "Systran/faster-whisper-large-v3"
      ],
      "speaches_container_mem_before_load": "8.847GiB / 121.7GiB",
      "speaches_container_mem_after_runs": "12.73GiB / 121.7GiB",
      "unload": "not requested; left to Speaches' idle TTL (DELETE /api/ps deadlocks this build)"
    },
    {
      "model": "Systran/faster-whisper-large-v3",
      "clip": "es",
      "median_processing_s": 43.88,
      "median_rtf": 1.218,
      "words": 104,
      "wer": 0.029,
      "wer_accents_stripped": 0.029,
      "runs": [
        {
          "run": 1,
          "processing_s": 45.291,
          "rtf": 1.258,
          "words": 104,
          "wer": 0.029,
          "wer_accents_stripped": 0.029,
          "text": "Bueno, empezamos. Hoy tenemos tres temas. Primero, el cierre del trimestre. Javier dice que 11 clientes todavía nos deben las facturas de septiembre, así que Pablo los llamará el martes y el miércoles. Segundo, las nuevas contrataciones. Lucía ha tramitado un alta en la seguridad social y está esperando el documento de identidad de la otra persona. Tercero, el asistente de inteligencia artificial. Ahorra tiempo en las respuestas sencillas, pero se equivoca con los importes, así que a partir de ahora nadie envía un correo con un importe sin que lo revise una persona. La próxima reunión es el lunes 19 a las 9."
        },
        {
          "run": 2,
          "processing_s": 43.875,
          "rtf": 1.218,
          "words": 104,
          "wer": 0.029,
          "wer_accents_stripped": 0.029,
          "text": "Bueno, empezamos. Hoy tenemos tres temas. Primero, el cierre del trimestre. Javier dice que 11 clientes todavía nos deben las facturas de septiembre, así que Pablo los llamará el martes y el miércoles. Segundo, las nuevas contrataciones. Lucía ha tramitado un alta en la seguridad social y está esperando el documento de identidad de la otra persona. Tercero, el asistente de inteligencia artificial. Ahorra tiempo en las respuestas sencillas, pero se equivoca con los importes, así que a partir de ahora nadie envía un correo con un importe sin que lo revise una persona. La próxima reunión es el lunes 19 a las 9."
        },
        {
          "run": 3,
          "processing_s": 43.826,
          "rtf": 1.217,
          "words": 104,
          "wer": 0.029,
          "wer_accents_stripped": 0.029,
          "text": "Bueno, empezamos. Hoy tenemos tres temas. Primero, el cierre del trimestre. Javier dice que 11 clientes todavía nos deben las facturas de septiembre, así que Pablo los llamará el martes y el miércoles. Segundo, las nuevas contrataciones. Lucía ha tramitado un alta en la seguridad social y está esperando el documento de identidad de la otra persona. Tercero, el asistente de inteligencia artificial. Ahorra tiempo en las respuestas sencillas, pero se equivoca con los importes, así que a partir de ahora nadie envía un correo con un importe sin que lo revise una persona. La próxima reunión es el lunes 19 a las 9."
        }
      ],
      "host_load_average_before": [
        3.62,
        3.32,
        2.48
      ],
      "host_load_average_after": [
        4.65,
        4.33,
        3.16
      ],
      "waited_for_load_s": 0,
      "speaches_loaded_models_after_runs": [
        "deepdml/faster-whisper-large-v3-turbo-ct2",
        "Systran/faster-whisper-large-v3"
      ],
      "speaches_container_mem_before_load": "8.847GiB / 121.7GiB",
      "speaches_container_mem_after_runs": "12.73GiB / 121.7GiB",
      "unload": "not requested; left to Speaches' idle TTL (DELETE /api/ps deadlocks this build)"
    }
  ],
  "load_average_note": "load_average_at_start is the first model's host_load_average_before: the script as run read the start value when it wrote the file (a bug fixed right after the run), so it duplicated the end value. Per-model before/after values were read at the right time."
}
